This document discusses artificial intelligence (AI) and its applications in biomedical fields. It begins by defining AI and biomedical AI as using algorithms and complex structures to analyze medical data similar to human intelligence. The document then discusses how AI is used in areas like medical imaging for tasks like cancer detection, as well as health monitoring, managing medical records, and diagnostics. It also explores technologies like machine learning and deep learning that power biomedical AI applications. Overall, the document provides a high-level overview of the evolution and uses of AI in healthcare and biomedical fields.
Vector Databases 101 - An introduction to the world of Vector Databases
AI in Biomedical: A Review of Key Applications and Challenges
1.
2. 4
ABSTRACT
Artificial Intelligence is the theory and development of computer systems that are able to
perform tasks that would require human intelligence. AI in healthcare is the use of algorithms
and software to approximate human cognition in the analysis of complex medical data. It
mainly refers to doctors and hospitals accessing vast data sets of potentially life- saving
information. This includes treatment methods and their outcomes, survival rates, and speed of
care gathered across millions of patients, geographical locations and innumerable and
sometimes interconnected health conditions.
Algorithms are already outperforming radiologists at spotting malignant tumours, and guiding
researchers in how to construct cohorts for costly clinical trials. Imaging, on the other hand
has become an essential component of many fields in medicine, biomedical applications,
biotechnology and laboratory research by which images are processed and analysed. Putting
together AI and imaging, the tools and techniques of artificial intelligence are useful for
solving many biomedical problems and using a computer based equipped hardware software
application for understanding images, researchers and clinicians can enhance their ability to
study, diagnose, monitor, understand and treat medical disorders.
AI is used in medical imaging to analyze breast cancer(Sonar ,MRI,CT), liver fibrosis and
tumour etc. Medical imaging is the technique and process of creating visual representation of
the interior of a body for clinical analysis. Cardiac CT is a painless imaging test that uses x
rays to take many detailed pictures of your heart and blood vessels, AI can provide insights by
processing the data and may even notice patterns that are not immediately obvious to the eye.
AI has an important role to play in the health care offerings of the future. Artificial Intelligence
will enable the next generation of radiology tools that are accurate and detailed enough to
replace the need for tissue samples in some cases. . There are a number of research studies
suggesting that AI can perform as well or better than humans at key healthcare tasks, such as
diagnosing disease.AI is not one technology , but rather a collection of them.
3. 5
INDEX
Chapter No Title Page no
Certificate 2
Acknowledgement 3
Abstract 4
1. Introduction
1.1.Need of AI
1.2.Potential of AI
7-9
2. Literature Survey 9
3. Evolution of AI in healthcare 10-11
4. Key Technologies
3.1.Machine Learning
3.2.Deep Learning
12-13
5. Use cases of AI within Biomedical
4.1. Medical Imaging
4.1.1. Tasks involved in Image Analysis
4.2.1. Challenges
4.2. Virtual Assistance
4.3. Health Monitoring
4.4. Managing Medical Records and Other Data
4.5. AI for Diagnostics
14-22
6. Applications currently in the Experimental phase
5.1. Radiology
5.2. Pain Monitoring
5.3. Melafind
23
7. Impact of AI in Biomedical
5.1. Trends in AI adoption
5.2. Impacts of AI on jobs
24-25
8. AI in the Global Healthcare market
6.1. Benefits of using AI in Healthcare
6.2. Risks and Challenges for AI in Healthcare
6.3. Possible solutions to deal with risks
25-29
9. Changes need to encourage the introduction and scaling
of AI in healthcare
29-30
4. 6
10. Future of AI in Biomedical 31
11. Conclusion 31
12. Related Research Papers 32-37
13. References 38
5. 7
CHAPTER 1: INTRODUCTION
The development of Artificial Intelligence (AI) in healthcare has been a long road with many
significant obstacles that at the same time present opportunities for biomedical engineers and
medical physicists to assume leadership roles in the implementation of AI in healthcare.
Artificial Intelligence or AI- Refers to the simulation of human intelligence in machines that
are programmed to think like humans and mimic their actions.
Biomedical-Is the application of engineering principles and design concepts to medicine and
biology concepts for healthcare purposes.
Artificial Intelligence (AI) in biomedical-
Its usage of software and complex structure of algorithms to mirror human intelligence in the
analysis of composite medical data. Specifically, Artificial Intelligence is the capability for
computer algorithms to estimate results without direct human interaction.
Since the first introduction of the concept in 1955, artificial intelligence (AI) has been a
“moving target” that always covered the most modern computing techniques aimed at
achieving things that were previously the exclusive task of humans.What distinguishes AI
technology from traditional technologies in health care is the ability to gain information,
process it and give a well-defined output to the end-user.All of these advances open questions
about how such capabilities can support,or even enhance, human decision making in health
and healthcare.AI does this through machine learning algorithms and deep learning.
The primary aim of health related AI applications is to analyze relationships between
prevention or treatment techniques and patient outcomes.the system deals with medical data
and knowledge domain in diagnosing patients conditions as well as recommending suitable
treatments for the particular patients. Major disease areas that use AI tools include
cancer,neurology and cardiology.The system serves to improve the quality of medical decision
making ,increase patients' compliance. AI in techniques in medical applications could reduce
the cost,time,human expertise and error.Due to the rapid development of AI software and
hardware technologies, AI has been applied in various technical fields mainly in biomedical
.This progress provides new opportunities and challenges as wells as directions for the future
of AI in biomedical.
The purpose of Artificial Intelligence is to make computers more useful in solving problematic
healthcare challenges and by using computers we can interpret data which is obtained by
diagnosis of various chronic diseases like Alzheimer, Diabetes, Cardiovascular diseases and
various types of cancers like breast cancer, colon cancer etc. It helps in early detection of
various chronic diseases which reduces economic burden and severity of disease.Even to
review the current state of AI in health,along with opportunities,challenges and practical
implications.
6. 8
Need of AI
● Computers are fundamentally well suited to performing mechanical computations
using fixed programmed rules.
● Artificial machines perform simple monotonous tasks efficiently and reliably,which
humans are ill-suited to.
● For more complex problems ,things get more difficult.Unlike humans,computers have
trouble understanding specific situations,and adapting to new situations.
● Artificial Intelligence aims to improve machines' behaviour in tackling such complex
tasks.
● Humans have an interesting approach to problem solving,based on abstract
thought,high-level deliberative reasoning and pattern recognition.
● AI research is allowing us to understand our intelligent behaviour.
● Artificial Intelligence can help us understand this process by recreating it, then
capability enabling us to enhance it beyond our current capabilities.
Potential of AI
AI has been around for decades and its promise to revolutionize our lives has been frequently
raised, with many of the promises remaining unfulfilled. Fueled by the growth of capabilities
in computational hardware and associated algorithm development, as well as some degree of
hype, AI research programs have ebbed and flowed. The JASON 2017 report gives this
history and also comments on the current AI revolution stating: “Starting around 2010, the
field of AI has been jolted by the broad and unforeseen successes of a specific, decades-old
technology: multi-layer neural networks (NNs). This phase-change reenergizing of a
particular area of AI is the result of two evolutionary developments that together crossed a
qualitative threshold: (i) fast hardware Graphics Processor Units (GPUs) allowing the
training of much larger—and especially deeper (i.e., more layers)—networks, and (ii) large
labeled data sets (images, web queries, social networks, etc.) that could be used as training
testbeds.
This combination has given rise to the “data-driven paradigm” of Deep Learning (DL) on
deep neural networks (DNNs), especially with an architecture termed Convolutional Neural
Networks (CNNs).” Is the current era just another hype cycle? Or are things different this
time that would make people receptive to embracing the promise of AI applications in health
and health care? AI is largely exciting to computational sciences researchers throughout
academia and industry. Perhaps previously the revolutionary advances in AI had no obvious
way to touch the lives of individuals.
The opportunities from health, including health care delivery, for AI may today be enhanced
by current societal factors that make the fate of AI hype different this time. Currently, there is
7. 9
great frustration in the cost and quality of care delivered by the US healthcare system. To
some degree, this has fundamentally eroded patient confidence, opening people’s minds to
new paradigms, tools, services. Dovetailing with this, there is an explosion in new personal
health monitoring technology through smart device platforms and internet-based interactions.
CHAPTER 2: LITERATURE SURVEY
Health Care Employees’ Perceptions of the Use of Artificial Intelligence Applications
Bahjat Fakieh, PhD Information Systems Department , King Abdulaziz University Al-
Solaimaniah District Jeddah.
The advancement of health care information technology and the emergence of artificial
intelligence has yielded tools to improve the quality of various health care processes. Few
studies have investigated employee perceptions of artificial intelligence implementation in
Saudi Arabia and the Arabian world. In addition, limited studies investigated the effect of
employee knowledge and job title on the perception of artificial intelligence implementation
in the workplace.
URL- https://ieeexplore.ieee.org/
Gudivada and N. Tabrizi, "A Literature Review on Machine Learning Based Medical
Information Retrieval Systems," 2018 IEEE Symposium Series on Computational
Intelligence (SSCI), Bangalore, India, 2018, pp. 250-257
As many fields progress with the assistance of cognitive computing, the field of health care is also
adapting, providing many benefits to all users. However, advancements in this area are hindered
by several challenges such as the void between user queries and the knowledge base, query
mismatches, and range of domain knowledge in users. In this paper, we present existing
methodologies as well as look into existing real-life applications that are used in the medical field
today.Future information retrieval (IR) models that can be tailored specifically for medically
intensive applications which can handle large amounts of data are explored as well.
URL-https://ieeexplore.ieee.org/
8. 10
CHAPTER 3: EVOLUTION OF AI IN HEALTHCARE
Among the many technology changes over the last decade we have seen the substantial growth
of data analytics for handling,processing,and gainfully using large amounts of
data.However,since data analytics can only work with historical data and give outcomes as
predefined by humans,specific rule based algorithms were developed to augment data
analytics,thereby imparting the self-learning capability to computer,which is now referred to
as “Machine Learning”.Machine Learning did not require the computers to be explicitly
programmed,which is a definitive advantage.Machine learning was then combined with data
analytics to analyze data and develop complex algorithms to predict models,which was named
as predictive analytics.
The evolution of AIS and its application has a vast spectrum in healthcare. The most important
reason is that there is non-availability of trained manpower in both medical and para-medical
fields. While Doctors, Nurses, Physiotherapists, dieticians and Lab/Imaging Technicians are
the front-end clinical staff, back-end human resources like Medical records technicians, billing
staff, Administrative staff, maintenance staff, marketing staff, Finance/accounting staff, IT
staff, etc. How can all these be replaced by AIS?
Let's take the work of doctors and nurses. AIS can have algorithms for arriving at a differential
diagnosis based on symptoms and further also advise investigations based on another set of
algorithms to arrive at a probable diagnosis. Even algorithms can be developed on likely
treatment. But let us not forget that Medicine is both an art and science. It is not a pure science;
hence two plus two will not make four. Not all diseases can be diagnosed on AIS based
algorithms and even treatments will differ based on individual genetic makeup. Example-
patients may have allergic reactions to certain drugs and react differently to the same set of
"factory based AIS output of treatments. There is this famous saying, medicine is almost the
same, the rest is in the mind!! The emotional and human touch of compassion of a nurse or a
doctor cannot be replaced ever by AIS based treatment modalities. The care taken by a nurse
cannot be replaced by AIS.
9. 11
The idea behind AIS in medicine is not so much to replace the but to enhance the doctor’s
medical expertise. A.I. programs take the amassed knowledge that every good physician has
which is the product of everything she learned in medical school and in training as well as her
experience in treating patient after patient and scale it to unprecedented levels. Why should
patients have access to just one particular doctor’s expertise when it’s now possible to provide
them with the brainpower of hundreds of thousands? Why should patients in rural areas who
live geographically far from the nation’s leading medical centers be deprived of all the up-to-
date knowledge housed there? The way artificial intelligence starts to really impact what’s
going on in health care is to be able to start cloning all the expert knowledge, so now all of a
sudden you get access to all types of care, anywhere.
And with the amount of data available to physicians today—from information about disease
symptoms to new drugs, interactions between different drugs and how different people treated
in the same way can have very different outcomes—the ability to access and digest information
is fast becoming a required skill. And it’s one that machine learning is uniquely designed to
master. Doctors are realizing that if they want to make sense of massive amounts of data,
machine learning is a way of allowing them to learn from that data.
In cancer care the application of AIS is tremendous. For human doctors to digest all this
information on cancers would be nearly impossible, given the demands on physicians’ time to
see patients and keep up to date on the latest advances in their field. The potential benefit of
having an AIS “doctor” on call at every cancer hospital, no matter how small, can’t be
overstated. People with rarer cancers have more confidence, since they now have the
institutional knowledge of leading experts in their field at their disposal.
CHAPTER 4: KEY TECHNOLOGIES
MACHINE LEARNING:
10. 12
The value of machine learning in healthcare is its ability to process huge data sets beyond the
scope of human capability, and then reliably convert analysis of that data into clinical
insights that aid physicians in planning and providing care, ultimately leading to better
outcomes, lower costs of care, and increased patient satisfaction.
Applied Machine Learning in Healthcare
Machine learning in medicine has recently made headlines. Google has developed a machine
learning algorithm to help identify cancerous tumors on mammograms. Stanford is using a
deep learning algorithm to identify skin cancer. A recent JAMA article reported the results of
a deep machine-learning algorithm that was able to diagnose diabetic retinopathy in retinal
images. It’s clear that machine learning puts another arrow in the quiver of clinical decision
making.Still, machine learning lends itself to some processes better than others.
Algorithms can provide immediate benefit to disciplines with processes that are reproducible
or standardized. Also, those with large image datasets, such as radiology, cardiology, and
pathology, are strong candidates. Machine learning can be trained to look at images, identify
abnormalities, and point to areas that need attention, thus improving the accuracy of all these
processes. Long term, machine learning will benefit the family practitioner or internist at the
bedside. Machine learning can offer an objective opinion to improve efficiency, reliability,
and accuracy.
The Ethics of Using Algorithms in Healthcare
It’s been said before that the best machine learning tool in healthcare is the doctor’s brain.
Could there be a tendency for physicians to view machine learning as an unwanted second
opinion? At one point, autoworkers feared that robotics would eliminate their jobs. Similarly,
there may be physicians who fear that machine learning is the beginning of a process that
could render them obsolete. But it’s the art of medicine that can never be replaced. Patients
will always need the human touch, and the caring and compassionate relationship with the
people who deliver care. Neither machine learning, nor any other future technologies in
medicine, will eliminate this, but will become tools that clinicians use to improve ongoing
care.The focus should be on how to use machine learning to augment patient care. For
example, if I’m testing a patient for cancer, then I want the highest-quality biopsy results I
can possibly get. A machine learning algorithm that can review the pathology slides and
assist the pathologist with a diagnosis, is valuable. If I can get the results in a fraction of the
time with an identical degree of accuracy, then, ultimately, this is going to improve patient
care and satisfaction.
11. 13
DEEP LEARNING:
Deep learning (also known as deep structured learning) is part of a broader family of machine
learning methods based on artificial neural networks with representation learning. Learning
can be supervised, semi-supervised or unsupervised.
Deep learning is an artificial intelligence (AI) function that imitates the workings of the human
brain in processing data and creating patterns for use in decision making. Deep learning is a
subset of machine learning in artificial intelligence that has networks capable of learning
unsupervised from data that is unstructured or unlabeled. Also known as deep neural learning
or deep neural network.
Deep learning provides the healthcare industry with the ability to analyze data at exceptional
speeds without compromising on accuracy. It’s not machine learning, nor is it AI, it’s an
elegant blend of both that uses a layered algorithmic architecture to sift through data at an
astonishing rate. The benefits of deep learning in healthcare are plentiful – fast, efficient,
accurate – but they don’t stop there. Even more benefits lie within the neural networks formed
by multiple layers of AI and ML and their ability to learn. Yes, the secret to deep learning’s
success is in the name – learning.
Deep learning uses mathematical models that are designed to operate a lot like the human brain.
The multiple layers of network and technology allow for computing capability that’s
unprecedented, and the ability to sift through vast quantities of data that would previously have
been lost, forgotten or missed. These deep learning networks can solve complex problems and
tease out strands of insight from reams of data that abound within the healthcare profession.
It’s a skillset that hasn’t gone unnoticed by the healthcare profession.
Deep learning in healthcare has already left its mark. Google has spent a significant amount of
time examining how deep learning models can be used to make predictions around hospitalized
patients, supporting clinicians in managing patient data and outcomes.
Deep learning in healthcare has already been seen in medical imaging solutions, chatbots that
can identify patterns in patient symptoms, deep learning algorithms that can identify specific
types of cancer, and imaging solutions that use deep learning to identify rare diseases or
specific types of pathology. Deep learning has been playing a fundamental role in providing
medical professionals with insights that allow them to identify issues early on, thereby
delivering far more personalized and relevant patient care.
12. 14
CHAPTER 5: CURRENT USE CASES OF AI WITHIN
BIOMEDICAL
● Medical Imaging-
Medical imaging refers to techniques and processes used to create images of various
parts of the human body for diagnostic and treatment purposes within digital health.
Imaging seeks to reveal internal structures hidden by skin and bones.Medical imaging
equipment are manufactured using technology from the semiconductor industry.They
include CMOS integrated circuit chips and power semiconductor devices.
Medical image analysis involves measurements in medical images, i.e., the extraction of
relevant quantitative information from the images.Manual measurements by human experts in
large 3D medical imaging datasets (in particular by radiologists in clinical practice) are not
only tedious and time-consuming and thus impractical in clinical routine, but also subject to
significant intra- and inter-observer variability, which undermines the significance of the
clinical findings derived from them. There is, therefore, great need for more efficient, reliable,
and well-validated automated or semi-automated methods for medical image analysis to enable
computer-aided image interpretation in routine clinical practice in a large number of
applications. Which information needs to be quantified from the images is of course highly
application specific.
While many applications in computer vision involve the detection or recognition of an object
in an image,whereby the precise geometry of the objects is often not relevant (e.g., image
classification, object recognition) or may be known a priori (e.g.machine vision), medical
13. 15
image analysis often concerns the quantification of specific geometric features of the objects
of interest (e.g., their position, extent, size, volume, shape, symmetry, etc.),the assessment of
anatomical changes over time(e.g., organ motion, tissue deformation, growth,lesion evolution,
atrophy, aging, etc.), or the detection and characterization of morphological variation between
subjects (e.g., normal versus abnormal development, genotype related variability, pathology,
etc.). The analysis of 3D shape and shape variability of anatomical objects in images is thus a
fundamental problem in medical image analysis. Apart from morphometry, quantification of
local or regional contrast or contrast differences is of interest in many applications, in particular
in functional imaging, such as fMRI,PET, or MR diffusion and perfusion imaging. Within the
wide variety of medical imaging applications, most image analysis problems involve a
combination of the following basic tasks:
1. Image Segmentation
Image segmentation involves the detection of the objects of interest in the image and defining
their boundaries, i.e., discriminating between the image voxels that belong to a particular object
and those that do not belong to the object. Image segmentation is a prerequisite for
quantification of the geometric properties of the object, in particular its volume or shape. Image
segmentation can be performed in different ways: boundary wise by delineating the contour or
surface of the object in one (2D) or multiple (3D) image slices; region-wise by grouping voxels
that are likely to belong to the same object into one or multiple regions; or voxel-wise by
assigning each voxel in the image as belonging to a particular object, tissue class, or
background. Class labels assigned to a voxel can be probabilistic, resulting in a soft or fuzzy
segmentation of the image.Accurate 3D segmentation of complex shaped objects in medical
images is usually complicated by the limited resolution of the images (leading to loss of detail
and contrast due to partial volume artifacts) and by the fact that the resolution is often not
isotropic (mostly multi-slice 2D instead of truly 3D acquisitions). Hence, interpolation is
usually needed to fill in the missing information in the data. In clinical practice, precise 3D
measurements (e.g., volumetry) may be too tedious and time-consuming, such that often a
simplified, approximate 2D or 1D analysis is used instead (e.g., for estimation of lesion size).
2. Image Registration
Image registration involves determining the spatial relationship between different images,
i.e.establishing spatial correspondences between images or image matching, in particular based
on the image content itself. Different images acquired at different time points (e.g., before and
after treatment), or with different modalities(e.g., CT, MRI, PET brain images), or even from
different subjects (e.g., diseased versus healthy)often contain complementary information that
has to be fused and analyzed jointly, preferably at the voxel level to make use of the full
resolution of the images. Image registration is needed to compensate for a priori unknown
differences in patient positioning in the scanner, for organ or tissue deformations between
different time points, or for anatomical variation between subjects. After proper registration,
14. 16
the images can be resampled onto a common geometric space and fused, i.e., spatially
corresponding voxels can be precisely overlaid, which drastically facilitates the joint analysis
of the images. In some cases,when deformations are ignorable, the registration solution can be
represented as an affine transformation matrix with a small number of parameters, but in
general a more complex transformation in the form of a locally flexible deformation field is
needed to accommodate for non-rigid distortions between the images.
3.Image Visualization
The information that is extracted from the imagesideally needs to be presented in the most
optimal way to support diagnosis and therapy planning,i.e., such that the correct interpretation
by the user of all relevant image data is maximally facilitated for a specific application. For
3D medical images, 2D multi-planar visualization is not well suited to assess structural
relationships within and between objects in 3D, for which true 3D visualization approaches are
to be preferred. To this end, either surface rendering or volume rendering can be applied.
Surface rendering assumes that a 3D segmentation of the objects of interest is available and
renders these within a 3D scene under specified lighting conditions by assigning material
properties to each surface or surface element that specify its specular and diffuse light
reflection,transmission, scattering, etc. Volume rendering instead renders the image voxels
directly by specifying suitable transfer functions that assign each voxel a color and opacity
depending on their intensity. While in principle volume rendering does not require a prior
segmentation of the objects of interest, in practice a prior segmentation of the image is often
applied such that the transfer functions can be made spatially dependent and object specific,
which allows to discriminate between voxels with similar intensity belonging to different
objects. In clinical applications such as image-based surgery planning or image-guided
intraoperative navigation, additional tools need to be provided to manipulate the objects in the
3D scene to add virtual objects to the scene or to fuse the virtual reality scene with real-world
images. While such augmented reality techniques can improve the integrated presentation of
all available information during an intervention their introduction in clinical practice is far
from trivial.
Image segmentation, registration, and visualization should not be seen as separate subproblems
in medical image analysis that can be addressed independently, each using a specific set of
strategies. On the contrary, they are usually intertwined and an optimal solution for a particular
image analysis problem can only be achieved by considering segmentation,registration, and
visualization jointly.
Challenges-
Medical image analysis is complicated by differ ent factors, in particular the complexity of
the data, the complexity of the objects of interest, and the complex validation.
15. 17
1.Complexity of medical data-
Medical images are typically 3D tomographic images. The 3D nature of the images provides
additional information, but also an additional dimension of complexity. Instead of processing
the data in 2D slice by slice, 3D processing is usually more effective as it allows to take spatial
relationships in all three dimensions into account, provided that the resolution of the data in-
plane and out-plane is comparable. Medical images are based on different physical principles
and the quantification of the images is complicated by the ambiguity that is induced by the
intrinsic limitations of the image acquisition process, in particular limited resolution, lack of
contrast, noise, and the presence of artifacts. Moreover, many applications involve the
analysis of complementary information provided by multiple images, for instance, to
correlate anatomical and functional information, to assess changes over time or differences
between subjects. It is clear that the variable, multi-X nature of the images to be analyzed
poses specific challenges.
2. Complexity of the Objects of Interest
The objects of interest in medical images are typically anatomical structures (sometimes also
other structures, e.g., implants), either normal or pathological (e.g., lesions), that can be rigid
(e.g., bony structures) or flexible to some extent (e.g., soft tissue organs). Anatomical
structures may exhibit complex shapes, such as the cortical surface of the brain, the cerebral
and coronary vessels, or the bronchial tree in the lung. Such complex shapes cannot easily be
described by a mathematical model. Moreover, anatomical structures can show large intra-
subject shape variability, due to internal soft tissue deformations (e.g), as well as inter-subject
variability, due to normal biological variation and pathological changes. In general, the
appearance of similar structures in different images (of the same subject at different time
points or from different subjects) can show significant variability, both in shape and in
intensity. Computational strategies for medical image analysis need to take this variability
into account and be sufficiently robust to perform well under a variety of conditions.
3. Complexity of the Validation
Medical image analysis involves the quantification of internal structures of interest in real-
world clinical images that are not readily accessible from the outside.Hence, assessment of
absolute accuracy is often impossible in most applications, due to lack of ground truth.As an
alternative, a known hardware phantom that mimics the relevant objects of interest could be
imaged, but the realism of such a phantom compared to the actual in vivo situation is often
questionable.Moreover,a hardware phantom usually constitutes a fairly rigid design that is
not well apt to be adapted to different variable anatomical instances. Instead, the use of a
software phantom in combination with a computational tool that generates simulated images
16. 18
based on a model of the imaging process provides more flexibility, with respect to both the
imaged scene and the image acquisition setup itself. But again, such simulated images often
fail to capture the full complexity of real data.
Examples-
1. Detection of diabetic retinopathy in retinal fundus images- Many diseases of the eye can
be diagnosed through non-invasive imaging of the retina through the pupil. Early screening for
diabetic retinopathy is important as early treatment can prevent vision loss and blindness in the
rapidly growing population of patients with diabetes. Such screening also provides the
opportunity to identify other eye diseases, as well as providing indicators of cardiovascular
disease. The increasing need for such screening, and the demands for expert analysis that it
creates, motivates the goal of low cost, quantitative retinal image analysis. Routine imaging
for screening uses the specially designed optics of a ‘fundus camera,’ with several images
taken at different orientations (fields, see Figure 2) and can be accomplished with (mydriatic)
or without (non-mydriatic) dilation of the pupil. Assessment of the image requires skilled
readers, and may be performed by remote specialists. With the advent of digital photography,
digital recording of retinal images can be carried out routinely through Picture Archiving and
Communication Systems (PACS).Figure 2: Standard image formats for diabetic retinopathy
(right eye). Source: taken from EYEPACS LLC 2017. As a point of reference, the standards
for screening for diabetic retinopathy in the UK require at least 80% sensitivity and 95%
specificity to determine referral for further evaluation. Screening using fundus photography,
followed by manual image analysis, yields sensitivity and specificity rates cited as 96%/89%
when two fields (angles of view) are included, and 92%/97% for three fields. (For a single
field, cited rates are 78%/86%).
17. 19
Recently a transformational advance in automated retinal image analysis, using Deep Learning
algorithms, has been demonstrated. The algorithm was trained against a data set of over
100,000 images, which were recorded with one field (macula-centered). Each image in the
training set was evaluated by 3-7 ophthalmologists, thus allowing training with significantly
reduced image analysis variability. The results from tests on two validation sets, also involving
only one image per eye (fovea centered), are striking. Selecting for high specificity (low false
negatives), yielded sensitivities/specificities of 90.3%/98.1% and 87.0%/98.5%). Selecting for
high sensitivity yielded values of 97.5%/93.4% and 96.1%/93.9%). These results compare
favorably with manual assessments even where those are based on images from multiple fields
as noted above. They also are a significant advance over previous automated assessments,
which consistently suffered from significantly lower sensitivities. The Deep Learning
algorithm shows great promise to provide increased quality of outcomes with increased
accessibility.
2. Dermatological classification of skin cancer- Skin cancer represents a challenging
diagnostic problem because only a small fraction (3–5% of about ~1.5 million annual US skin
cancer cases) are the most serious type, melanoma, which accounts for 75% of the skin cancer
deaths. Identifying melanomas early is a critical health issue, and because diagnosis can be
performed on photographic images, there are already services that allow individuals to send
their smart-phone photos in for analysis by a dermatologist. However, the detection of
melanomas in screening exams is limited – sensitivity 40.2% and specificity 86.1% for primary
care physicians and 49.0%/ 97.6% for dermatologists.A recent demonstration of automated
skin cancer evaluation using a convolutional neural network (CNN) algorithm yielded striking
results.The authors drew on a training set of over 125,000 dermatologists labeled images, from
18 different online repositories. Two thousand of the images were also labeled based on
biopsies. The algorithm was trained on all the dermatologist labeled images, using 757 disease
classes and over 2000 diseases.
18. 20
1. A further classification test was performed drawing only on images that were biopsy-
proven to be in a specific disease class. The algorithm then was run to answer only the
question of whether the lesion in the image was benign or malignant. The results for
analysis of 130 images of melanocytic lesions are shown in Figure 3b, compared with
results from assessments by 22 different dermatologists. As with the broader
classification tests, the algorithm performs similarly or slightly better than individual
dermatologists. The performance for both algorithms and dermatologists is much better
for this specific task than for the classification, noted above, of images from a set
representing all the different diseases. As with the retinopathy example, these results
indicate that AI algorithms can perform at levels matching their training sets. The poor
level of results for the broad screening tests is consistent with the training set, which is
based on dermatological characterization.
More use cases are-
1.Virtual health assistants:
Using augmented reality, cognitive computing, speech and body recognition software, a
virtual persona is created for patients to engage with. These virtual health assistants are able
to provide a personalized experience in which patients can ask questions and learn how to
better manage their health.
19. 21
2.Health Monitoring:
Wearable health trackers – like those from FitBit, Apple, Garmin and others – monitors heart
rate and activity levels. They can send alerts to the user to get more exercise and can share this
information to doctors (and AI systems) for additional data points on the needs and habits of
patients.
3. Managing Medical Records and Other Data:
Robots collect, store, re-format, and trace data to provide faster, more consistent access.E.g.
Nuance is a production service provider that uses AI and machine learning in order to predict
a particular user's intent and by implementing nuance in an organisation workflow you can
develop a personalized user experience that allows the company to make better decisions and
better action.
Nuance provides AI powered solutions to help doctors cut documentation time and improve
reporting quality.nuance basically helps in storing,collecting and reformatting data in order to
provide faster and more consistent access to all the data so that any further analysis or any
diagnosis.
4. AI for Diagnostics:
Determining a patient’s diagnosis is a vital aspect of healthcare. Care providers and medical
researchers alike can see the useful potential of using AI to augment or replace the human
ability to identify illness and disease.
5. Mental and Physical Health Screening
Another aspect of healthcare that is primed for assistive AI—in some cases, already seeing the
use of AI—is the diagnostic screening process. This is typically performed by a patient
speaking with a doctor or other healthcare professional and answering a series of questions
about their medical history and describing symptoms, which the health care provider uses to
make a diagnosis or recommend a course of action for the patient.
6. Industry
The use of AI in the healthcare industry section could be another article altogether, such vast
and detailed are the applications. The main focus of AI in the health sector is in the clinical
decision support systems, several industrial Giants are widely in use of these systems.
20. 22
1. Microsoft’s Hanover project, in partnership with Oregon Health and Science
University’s Knight Cancer Institute, analyzed medical research to predict the most
effective cancer drug treatment options for patients.
2. Google’s DeepMind is being used by the UK National Health Service to detect certain
health risks through data collected with the use of a mobile app.
3. Apple Iphone’s health app keeps track of users’ activities and he can check and analyze
his lifestyle, monitoring heart rate Pulse, and steps taken in a day.
CHAPTER 6: CURRENT RESEARCH
Various fields of medicine have inculcated AI into their procedures and achieved
improvement.
● Radiology- This field has grasped the most popular so far, having acquired the
adeptness to interpret imaging results and detecting minute changes which could
otherwise be easily missed by the human eye. Algorithms with higher resolution have
been implemented to detect diseases like pneumonia with better accuracy.
● Pain management: This is still an emergent focus area in healthcare. As it turns
out, by leveraging virtual reality combined with artificial intelligence, we can create
simulated realities that can distract patients from the current source of their pain and
even help with the opioid crisis.
● MelaFind: This technology uses infrared light to evaluate pigmented lesions. Using
algorithms, dermatologists can analyze irregular moles and diagnose serious skin
cancers such as melanoma. Although this technology should not replace a biopsy, it
helps with giving an early identification, Dr. Weber said.
CHAPTER 7: IMPACT OF AI IN BIOMEDICAL
Technology has already improved diagnostic accuracy, drug delivery, and patients’ medical
records, and AI will only add to those breakthroughs. AI can mine medical records, design
personalized treatment plans, handle administrative tasks to free up medical providers’ time
for more meaningful tasks, and assist with medication management.
21. 23
AI has already made headway in medicine, helping to do everything from processing x-ray
images and detecting cancer to assisting doctors in diagnosing and treating patients. In fact,
the global AI healthcare market is expected to reach $22,790 million by 2023.And the general
public is on board. According to a recent survey, 47% of people were comfortable with AI
assisting doctors in the operating room. More than half of respondents over age 40 were willing
to go under the knife with the help of technology, compared with only 40% under age 40.
Additionally, six in ten participants (61%) were comfortable with their doctor using data from
wearable devices, such as an Apple Watch or Fitbit, to assess their lifestyle and make
recommendations based on that data.
So what healthcare areas will AI have an impact on in the next five to ten years?
● Mining medical records
In our current age of big data, patient data is valuable. Oftentimes, patients’ files are
unorganized and mining their records to extract necessary medical insights can be a great
challenge.E.g.David Lindsay, founder of Philadelphia-based start-up, Oncora Medical,
realized this struggle in radiation therapy. He and his team built a data analytics platform that
helps doctors design sound radiation treatment plans for patients, personalizing each one based
on their specific characteristics and medical history.
● Drug development
Clinical trials can take more than a decade and cost millions of dollars. AI can play a part in
speeding up the process of drug development, along with making it more cost effective.GSK,
a company that researches, develops, and manufactures innovative pharmaceutical medicines,
vaccines, and consumer healthcare products, is actively applying AI to its drug discovery arm.
In fact, it created an in-house AI unit called “Medicines Discovered Using Artificial
Intelligence.” In 2017, the company announced a partnership with Insilico, to identify novel
biological targets and pathways.
● On jobs of Healthcare sector
Emergence of AI in healthcare has instigated a fear among people about losing jobs,eventually
slowing down the adoption of AI among healthcare workers.most federal governments and
policy makers have a misconception that with increasing adoption of AI,jons would become
redundant thus adversely affecting the economic goal of job creation.
22. 24
on the contrary,it is being analysed that with adoption of AI,the employment opportunities are
going to increase and new age skills would be in great demand.Many jobs like caregiving and
rehabilitation require human emotions and utmost care which AI cannot currently replicate.AI
is integrated in healthcare organisations to assist with care provisions,not replace
it.moreover,as AI continues to evolve in healthcare,there would be more job created for new
skills sets.Ai in healthcare would have advantages of increased efficiency and decreased costs
of treatment,leading to higher profit and employment opportunities.Thus,it is a misconceptions
that AI would replace a healthcare workers in reality it can lead to an increase in demand of a
qualified workforce and improve efficiency in services like diagnostics,patients,engagement
and precision medicine.
CHAPTER 8: AI IN THE GLOBAL HEALTHCARE MARKET
Similar to other industries,healthcare is witnessing a shift to consumerization,pushing payers
and providers to focus on value based care and improve the health outcomes.Across various
geographics advanced tools like ai are being implemented to address varied stakeholders
challenges and augmented care provision,in most economics,irrespective of the stages of
development,the cost and demand for care is rising,there by increasing the need for digital
technologies,it becomes imperative to provide seamless and integrated care by leveraging the
benefits of the connected ecosystem where patients providers,payers and other stakeholders
are increasingly adopting technology to simplify the processes.
Advanced and developed economies like US,Germany,canada and UK spend a huge
proportion of gdp on healthcare however the adoption of proven technologies like AI is yet to
gain importance in their health system.Through US is the highest spender on healthcare
23. 25
globally as a percentage of its GDP,it faces challenges like rising cost of healthcare
provision,storage of primary care professions,poor quality outcome and lack of coverage fora
high percentage of the population.US spends two and a half times higher than the average of
organisation for economic co-operation and development(OECD)countries on healthcare,with
significant proportion being out of pocket or voluntary coverage.it also has the highest rates of
medication errors compared to other OECD countries.The average insurance subscription in
US is about USD400 a month and significant amount of healthcare service contributions are
co-payments.
Germany AI in Healthcare Market Size, Share & Trends Analysis Report by Offering
(Hardware and Software & Services), By End-User Industry (Hospitals & Healthcare
Facilities, Personal Care, Biotechnology & Pharmaceutical Companies), By Application
(Diagnosis, Biomarker, Virtual Nursing Assistance, Remote Monitoring of Patients, Drug
Discovery, and AI-Enabled Hospital Care), and Forecast 2019-2025.
Germany AI in the healthcare market is estimated to grow significantly at a CAGR of 51.6%
during the forecast period. The presence of well-established and start-up companies is one of
the major factors driving the growth of the AI in the healthcare market in the country. The
market is segmented on the basis of offering, end-user industry, and application. Based on
offering, the market is divided into software & services and hardware. Based on the end-user
industry, the market is segmented into hospitals & healthcare facilities, personal care, and
biotechnology & pharmaceutical companies. Further, on the basis of application, the market is
segmented into diagnosis, biomarker, virtual nursing assistants, remote monitoring of patients,
drug discovery, and ai-enabled hospital care.
BENEFITS
1.Job stability: According to the United States Bureau of Labor Statistics, the healthcare
industry is projected to grow 18 percent from now until 2026, much faster than the average for
all occupations. This projected growth is mainly due to an aging population and a greater
demand for healthcare services. Plus, it doesn’t matter where you are in the world, there will
always be people in need of help. In a shaky economy and world of uncertainty, having this
much job security is a huge advantage.
2.Great pay and benefits: As of May 2017, the median annual wage for healthcare
practitioners and technical occupations (such as registered nurses, physicians and surgeons,
and dental hygienists) was $64,770 – almost double the median annual wage for all
occupations. Typically, the more training you have, the better the wages will be. For example,
the average base pay for a neurosurgeon is $489,839 per year.
24. 26
3.Fast-paced workday: It’s likely that your career in healthcare will be highly stimulating
with a constantly changing atmosphere (bye, bye 9-5 desk job). What your workday looks like
depends on your specialty but be prepared to work face-to-face with patients and be on your
feet most of the day. The medical field is full of excitement, and you’ll never live the same day
twice.
4.Opportunities for growth: You don’t need years of medical training to make a difference
in someone’s life. Some specialties only require a certificate, which could be achieved in a
year or two. Plus, medical facilities are looking for people to work in all areas of care, like
reception and administration. If you’re looking to work your way up, many companies also
offer continued learning programs and tuition reimbursement.
5.The chance to help people: Those who work in the healthcare industry typically have a
desire to make a difference. Whether you’re the surgeon who removes debilitating tumors or
the receptionist who offers a friendly smile to a patient who just received a difficult diagnosis,
you’re there for patients and families when they need it most.
6.AI helps in early diagnosis: By implementing AI, healthcare professionals can reap the
benefit of early detection by pinpointing any risks highlighted by the AI algorithm. The AI
database gathered over a period of time compiles a lot of symptoms and diagnosis to accurately
predict potential health risks in a patient.
7.AI cuts down time needed in diagnosis: AI based healthcare apps have a strong advantage
in coming up with accurate disease diagnosis in a swift time frame. This is possible because of
the amount of data and millions of symptoms/diagnosis these AI apps have. This makes AI
more time efficient and cost efficient in coming up with the disease diagnosis.
CHALLENGES AND RISKS
1.Injuries and errors—The most obvious risk is that AI systems will sometimes be wrong,
and that patient injury or other health-care problems may result. If an AI system recommends
the wrong drug for a patient, fails to notice a tumor on a radiological scan, or allocates a
hospital bed to one patient over another because it predicted wrongly which patient would
benefit more, the patient could be injured. Of course, many injuries occur due to medical error
in the health-care system today, even without the involvement of AI. AI errors are potentially
different for at least two reasons. First, patients and providers may react differently to injuries
25. 27
resulting from software than from human error. Second, if AI systems become widespread, an
underlying problem in one AI system might result in injuries to thousands of patients—rather
than the limited number of patients injured by any single provider’s error.
2.Data availability—Training AI systems requires large amounts of data from sources such
as electronic health records, pharmacy records, insurance claims records, or consumer-
generated information like fitness trackers or purchasing history. But health data are often
problematic. Data is typically fragmented across many different systems. Even aside from the
variety just mentioned, patients typically see different providers and switch insurance
companies, leading to data split in multiple systems and multiple formats. This fragmentation
increases the risk of error, decreases the comprehensiveness of datasets, and increases the
expense of gathering data—which also limits the types of entities that can develop effective
health-care AI.
3.Privacy concerns—Another set of risks arise around privacy.The requirement of large
datasets creates incentives for developers to collect such data from many patients. Some
patients may be concerned that this collection may violate their privacy, and lawsuits have
been filed based on data-sharing between large health systems and AI developers. AI could
implicate privacy in another way: AI can predict private information about patients even
though the algorithm never received that information. For instance, an AI system might be able
to identify that a person has Parkinson’s disease based on the trembling of a computer mouse,
even if the person had never revealed that information to anyone else. Patients might consider
this a violation of their privacy, especially if the AI system’s inference were available to third
parties, such as banks or life insurance companies.
4.Distributional shift — A mismatch in data due to a change of environment or
circumstance can result in erroneous predictions. For example, over time, disease patterns
can change, leading to a disparity between training and operational data.
5.Reinforcement of outmoded practice — AI can’t adapt when developments or changes in
medical policy are implemented, as these systems are trained using historical data.
6.Self-fulfilling prediction — An AI machine trained to detect a certain illness may lean
toward the outcome it is designed to detect.
7.Negative side effects — AI systems may suggest a treatment but fail to consider any
potential unintended consequences.
26. 28
8.Unsafe exploration — In order to learn new strategies or get the outcome it is searching
for, an AI system may start to test boundaries in an unsafe way.
9.Unscalable oversight — Because AI systems are capable of carrying out countless jobs
and activities, including multitasking, monitoring such a machine can be near impossible.
POSSIBLE SOLUTIONS
There are several ways we can deal with possible risks of health-care AI:
Data generation and availability-Several risks arise from the difficulty of assembling high-
quality data in a manner consistent with protecting patient privacy. One set of potential
solutions turns on government provision of infrastructural resources for data, ranging from
setting standards for electronic health records to directly providing technical support for high-
quality data-gathering efforts in health systems that otherwise lack those resources. A parallel
option is direct investment in the creation of high-quality datasets.
Quality oversight- Oversight of AI-system quality will help address the risk of patient injury.
The Food and Drug Administration (FDA) oversees some health-care AI products that are
commercially marketed. The agency has already cleared several products for market entry, and
it is thinking creatively about how best to oversee AI systems in health. However, many AI
systems in health care will not fall under FDA’s purview, either because they do not perform
medical functions or because they are developed and deployed in-house at health systems
themselve a category of products the FDA typically does not oversee. These health-care AI
systems fall into something of an oversight gap. Increased oversight efforts by health systems
and hospitals, professional organizations like the American College of Radiology and the
American Medical Association, or insurers may be necessary to ensure quality of systems that
fall outside the FDA’s exercise of regulatory authority.
27. 29
CHAPTER 9: CHANGES NEED TO ENCOURAGE THE
INTRODUCTION AND SCALING OF AI IN HEALTHCARE
The strides made in the field of AI in healthcare have been momentous. Moving to a world in
which AI can deliver significant, consistent, and global improvements in care will be more
challenging.
Of course, AI is not a panacea for healthcare systems, and it comes with strings attached. The
analyses in this report and the latest views from stakeholders and frontline staff reveal a set
of themes that all players in the healthcare ecosystem will need to address:
What needs to change to encourage the introduction and scaling of AI in healthcare?
The strides made in the field of AI in healthcare have been momentous. Moving to a world in
which AI can deliver significant, consistent, and global improvements in care will be more
challenging.
Of course, AI is not a panacea for healthcare systems, and it comes with strings attached. The
analyses in this report and the latest views from stakeholders and frontline staff reveal a set
of themes that all players in the healthcare ecosystem will need to address:
1. Working together to deliver quality AI in healthcare. Quality came up in our
interviews time and again, especially issues around the poor choice of use cases, AI
design and ease of use, the quality and performance of algorithms, and the robustness
and completeness of underlying data. The lack of multidisciplinary development and
early involvement of healthcare staff, and limited iteration by joint AI and healthcare
teams were cited as major barriers to addressing quality issues early on and adopting
solutions at scale. The survey revealed this is driven by both sides: only 14 percent of
startup executives felt that the input of healthcare professionals was critical in the
early design phase; while the healthcare professionals saw the private sector’s role in
areas such as aggregating or analyzing data, providing a secure space for data lakes,
or helping upskill healthcare staff as minimal or nonexistent.
One problem AI solutions face is building the clinical evidence of quality and
effectiveness. While startups are interested in scaling solutions fast, healthcare
practitioners must have proof that any new idea will “do no harm” before it comes
anywhere near a patient. Practitioners also want to understand how it works, where
the underlying data come from and what biases might be embedded in the algorithms,
28. 30
so are interested in going past the concept of AI as a “black box” to understand what
underpins it. Transparency and collaboration between innovators and practitioners
will be key in scaling AI in European healthcare.
User-centric design is another essential component of a quality product. Design
should have the end user at its heart. This means AI should fit seamlessly with the
workflow of decision makers and by being used, it will be improved. Many
interviewees agreed that if AI design delivers value to end users, those users are more
likely to pay attention to the quality of data they contribute, thereby improving the AI
and creating a virtuous circle. Finally, AI research needs to heavily emphasize
explainable, causal, and ethical AI, which could be a key driver of adoption.
2. Rethinking education and skills. We have already touched on the importance of
digital skills—these are not part of most practitioners’ arsenal today. AI in healthcare
will require leaders well-versed in both biomedical and data science. There have been
recent moves to train students in the science where medicine, biology, and
informatics meet through joint degrees, though this is less prevalent in Europe. More
broadly, skills such as basic digital literacy, the fundamentals of genomics, AI, and
machine learning need to become mainstream for all practitioners, supplemented by
critical-thinking skills and the development of a continuous-learning mind-set.
Alongside upgrading clinical training, healthcare systems need to think about the
existing workforce and provide ongoing learning, while practitioners need the time
and incentive to continue learning.
3. Strengthening data quality, governance, security and interoperability. Both
interviewees and survey respondents emphasized that data access, quality, and
availability were potential roadblocks. The data challenge breaks down into digitizing
health to generate the data, collecting the data, and setting up the governance around
data management. MGI analyses show that healthcare is among the least digitized
sectors in Europe, lagging behind in digital business processes, digital spend per
worker, digital capital deepening, and the digitization of work and processes. It is
critical to get the basic digitization of systems and data in place before embarking on
AI deployments—not least because the frustrations staff have with basic digitization
could spill over to the wider introduction of AI.
29. 31
CHAPTER 10: FUTURE OF AI IN BIOMEDICAL
Artificial Intelligence will dominate the healthcare industry in the future that can sense or
predict a disease outbreak(pandemic) from stopping it’s early outspread. In the early decades,
there was a time where people lacked medical services and technology on a disease outbreak.
The deaths were innumerable and diseases were unknown. Today’s scenario is devastating as
the world has witnessed a pandemic like any day before, besides highly equipped technology
and medical services.
This isn’t enough! The world requires more! Something that can change the whole situation
upside down. We have to shift to the advanced stage of healthcare where AI comes into the
rescue. Artificial Intelligence is about conquering human intelligence with its utmost
systemized configurations that can boost the healthcare industry to serve better and more at the
same time.
Today, many world-famous AI healthcare companies use this technology for the development
of cutting-edge solutions. And the major market players are still very familiar: IBM, Microsoft,
and Google. Here is just a overview of what they are working on:
● IBM applies AI to create solutions for cancer and chronic diseases treatment as well as
for the development of new medications;
● Microsoft conducts in-depth research on how AI can help predict cancer treatment
reactions and develops programmable cells;
● Google creates a platform that detects health risks for patients based on the mobile
software collected data
CHAPTER 11: CONCLUSION
As we take stock of how far AI has come, and how it has driven advances in digital health
technology, it’s easy to be excited by the future. Many questions still remain—how to
preserve the security and privacy of medical data, for example, or the unexpected hazards of
constant biomedical surveillance.
But with the prevalence of smartphones, wearable devices, AI assistants, and autonomous
robots, all of them brimming with medical applications, the future of digital health looks bright.
AI integrations have the potential to detect diseases earlier, track epidemics more effectively,
laser-target treatment options, and connect patients to their doctors in ways that a pre-
smartphone generation never thought possible.
30. 32
RESEARCH PAPERS
Uses of Artificial Intelligence in Health
Ed Godber,Chief Scientist,H-Labs.London, UK.
Abstract—The use of artificial intelligence in health is
growing rapidly. In this paper, a number of case
studies were reviewed to elicit the sensitivity of health
impact to parameters of AI design and application
context. At a high level, the speed at which
breakthroughs are happening is highly encouraging.
Whereas application to core science or health
preservation is working well, however, things are
much more challenging at the intersection with
healthcare itself. In particular lack of game alignment,
lack of regulation and cognitive dissonance between
the intuition of the AI architect and that of the
healthcare practitioner would benefit from much more
attention and cross fertilisation. Whilst not quantified
here, there were early signals that impact will be
highest where such exchange of intuition has been
explored and engineered into the system.
Keywords—Artificial Intelligence, Human Intuition,
Health, Regulation
I. Intelligence and knowledge in healthcare
Chess and Go, driverless cars and image recognition
have the advantage that the rules are certain, the
objects are known with certainty and it is possible to
generate millions of reliable data points. In areas of
health where data is reliable and the rules of biology
are well understood, algorithms are being introduced
with increasing frequency and success. Given the
capacity for machine learning to surpass such
algorithms, we are seeing ever more ambitious
endeavours spring up, before the first wave of
algorithmic technology has even been digested, and a
surge of investor interest in anything ‘AI’. In the health
sector, in order to protect the consumer and instil trust
in the system, there is comprehensive regulatory
oversight of the technology sector and professionals
are required to invest in significant training and
practice within guidelines to maintain certification.
Over time, this has led to a public culture and working
presumption that healthcare is built upon a foundation
of highly reliable knowledge and robust data. So, one
can understand why people might think that the limits
to human computational capacity to process all this
data and knowledge might be the bottleneck and AI
the solution. Unfortunately, the knowledge chain and
datasets in health have critical flaws at many points
and so experts are forever making decisions on the
basis of ‘least bad’ knowledge and adding their own
intuition (combining common sense with case
experience). In these scenarios, the direct insertion of
AI,without interoperability with human intuition, will
have unpredictable consequences.
Moreover, it is always a human who sits behind the
choice as to where to apply AI, how to generate the
learning model, what data to get, how to evaluate and
how to integrate with humans. But where do they get
the data from to make such choices and predict the
ripple effects over time? How well are they doing their
job and are they subject to certification too? In this
article, a broad range of applications of AI in health
are described, and examined with the uncertainty
context in mind. Using case studies, the goal is to
break down the AI versus human debate into
something less binary, insight based and useful for
generating evidence-led policy in the future.
II. Taking a structured approach to reviewing the
application of AI to health
For any planned application of AI to health, it would
be useful for commissioners, regulators, users and
practitioners to have a means of learning from other
projects. A database of publicly available case studies
that provide sufficient information on what problem
was being addressed, how, what type of AI model was
being used and some initial insight into
performance/issues would be a good starting place.
But it would ideally be codified in such a way that
related to key features that relate to impact and
risks.The aim of the research was to formulate an
empirically- driven hypothesis around parameters that
should be considered for such a search resource. The
method used was to identify case study applications
across four domains (preservation, surveillance, high-
end interventions and core science) which met the
‘sufficient information’ criteria listed above and to
pick out the key drivers/issues for each one. It was not
intended to be a formalised literature review or
qualitative analysis, but instead to guide that next
phase of research. For the purposes of generating a
strawman taxonomy, a brief description of 11 out of
the 18 case studies, together with the key learnings are
reported here, because the remaining 7 simply
reinforced the same themes.
III. Case Studies
IIIa. Preservation of health
Companies, such as Viome, apply supervised and
unsupervised machine learning to discover new ways
in which the biome links to preservation of health.
Core science is uncovering a strong relationship
between dysbiosis and health,but, if we want to be
preventing ill-health rather than treating disease, we
need to go much further. So, this is about
revolutionising the knowledge base and using it to
deliver personalised lifestyle changes to prolong
health.There is a circularity challenge to overcome, to
31. 33
some extent. One needs the individual data to discover
with certainty what the relevance of each bacterial type
is, but one needs to be able to provide insight and
recommendations that work in order to attract the
individual to give the data. Prior to the knowledge
reaching its disruptive state, the risk that only the most
bacteria-curious people might use the service, leading
to a potential bias in data. There will be challenges to
be overcome in sampling error and contamination,
integration of human expertise and cost. However, the
counter-factual is human trial and error based on a
longitudinal dataset of n=1 and the potential benefit
for health and science is very high. The Game' is a
combination of big data analytics and
personalisation, for which there is plenty of evidence
that this form of machine learning is appropriate.Other
enterprises are trying to do a similar thing, but using
deep learning to solve for a broader range of biological
variables (gene, RNA, protein, biome, clinical,
wearable,imaging, symptoms) and multiple forms of
health preservation and disease. A challenge such as
this has all the complexities of the biome case study,
but also has to overcome the measurement and
sampling issues in fields such as proteomics and
symptoms reporting, and the inter-relationships
between these biological parameters. The data
requirement is extraordinarily high and may rely on
sharing of sensitive health data on a regular basis by
many. In fields such as nutrition,exercise and skin
status, progress looks to be advancing quickly.
However, one would surmise that the medical field
may be more difficult because of the complexity of
phenotypic and proteomic data, cost of analysis and
data sensitivity is much higher there. Moreover, the
status quo is much more sophisticated than the
consumer’s trial and error process. The potential
contribution to knowledge, and the translation of that
into actions that can preserve health, is very large, but
one may need to be quite selective in what to
prioritise.The key takeaways from the application of
AI to health preservation are that: They are trying to
address one of the key problems in health – a lack of
knowledge. Whilst they replace the consumer’s ‘trial
and error’ activity in discovery, they transform the
consumer’s collection of data and focus/activity
around the way in which their behaviour affects the
emergent drivers of health. So, it re-directs the role of
the human rather than getting rid of it. The AI models
seem appropriate and the ‘games’ fit the nature of what
the real-world system is going to look like (as it is
being built around this AI model); but that, as the
sensitivity of data and complexity of measurement
increases (as get closer to disease) the challenge of
getting enough of the right data and making sure it is
used in a way in which the donor is comfortable with
becomes exponentially hard.
IIIb. Health Surveillance
Amongst the most common transitions from good
health to disease is the path towards heart attacks,
diabetes and complications related to diabetes. Work
carried out by DeepMind's AI platform established
that it was possible to get all that risk information from
a retinal fundus camera read-out2.Their discovery was
based on analysis of data from 284,355 patients. This
is important because, particularly in relation to
diabetes, such information is not gathered routinely in
a timely way in the real world. This is another example
of using AI to add to knowledge and quality of
measurement, this time for risk factors which are
already known – but to discover a new modality for
surveillance.Others have gone even further in respect
of diabetic retinopathy3, getting to impressive
sensitivity rate (87%).However, on an intention-to-
screen basis (all other technologies must demonstrate
significance on an intention-to treat basis) the level of
confidence went down to a level whereby superiority
disappeared (80%), because a large number of scans
were not of sufficient quality for the AI to do its job.
Moreover, this model is not designed to detect serious
eye problems such as age-related macular
degeneration or retinal detachment.These AI bots do
not get exposed to hundreds of thousands of visual and
human-to-human conversational cues, but the
physician does.It is also not clear how the AI is trying
to fit into the existing diagnostic process. Not having
the certification to operate as an independent
diagnostic provider and to activate a chosen clinical
strategy thereafter, if the individual does have a
medical problem they will end up having to repeat the
whole process again, but with a human diagnostic
model.For patients who accurately self-assess the need
to visit the physician (true positive) or not to go (true
negative), this AI the model adds to costs and might
confuse the patient. For patients who have a medical
condition but don’t go to the physician(false negative),
the question is whether the AI output changes the
behaviour of the patient and the physician gets it right.
For those that go to the doctor when they are actually
fine (false positive), will the AI persuade them not to
visit. Ironically, in a pilot scheme for replacing
telephone-based triage for emergency services with an
AI bot, there was anecdotally an increase in false
positives because people wouldn’t risk relying on the
AI bot. So, the decision to design this as running
independently of the physician-diagnostic process,
rather than explicitly integrated, has very important
implications. Notably, there are other AI systems
emerging that look to integrate more explicitly into
physician processes.Another application of AI is in
diagnostics which is already high-end and highly
specialised, such as detection of breast cancer through
mammography. Current practice often involves
getting the opinion of two different radiologists, which
creates shortages and bottlenecks. One AI company,
Kheiron Medical,secured the first ever CE approval
for a radiology AI device in Europe on the basis of a
study involving approximately 5,000 patient
mammograms in which it performed better than the
benchmark radiologist.. The ‘games’ AI were playing
required either a change in where diagnosis of cardiac-
risk takes place (in a more expensive setting) or to
improve detection of diabetes-related eye risks at the
expense of other types of eye-health risks. For the
general medical diagnosis model, it might seem more
logical to let human intuition optimise the linguistic
process by which data is gathered from the patient,
32. 34
iterating with an AI device to explore the medical
knowledge database more quickly and accurately, and
then let the expert integrate visual data and the
decision to gain further diagnostic clues through
performing tests (temperature, blood pressure,
physical examination, blood tests, urine tests etc.) and
then to activate a medical pathway. Some AI models
look a little more like this in their design.The
mammography case study sets somewhat of a gold
standard. They are essentially getting different data
and playing a different game and not acknowledging
the incompleteness of the medical taxonomy for many
areas, such as mental health.Such models require a
general consensus that, for better or worse, a different
game should be played that may create unpredictable
levels and types of error but deliver lower cost.That
latter point assumes monopoly power would not end
up being exploited into pricing well above the cost of
a physician.
III c. High-end interventions
One of the most technically challenging and risky
interventions in healthcare is transplantation or
regenerative therapy using stem cells. In an era of 3D
bioprinting, AI imaging technology can be very useful
for designing the precise blueprint for the organ. But
AI is being used in fascinating ways with respect to
stem cell science too.Nobel prize winner, Shinya
Yamanaka has been using AI to deal with one of the
big issues facing the use of induced pluripotent stem
(IPS) cells in regenerative medicine. There is a fear
that, due to genetic mutation, IPS cells may later turn
out to be cancerous.There is complete alignment with
the game’ that the regenerative process would entail
and it would solve the problem of important
knowledge about safety not being processed to the
point of care. Interestingly, Mahayo Takahashi, the
pioneer of IPS cell transplantation to regenerate
retinas, is working through Innovation Japan to use AI
to train robots to undertake the very precise and
complex procedure of creating IPS cell sheets. It may
take a little longer than training a scientist, but once
trained, that knowledge can be downloaded to other
robots at scale, rather than being constrained by a
modest train-the-trainer growth dynamic. This
approach is being used in robotic surgery too.The
striking thing about high-end interventions is that they
typically involve the leading experts using AI to solve
precisely the right problems, ones which too few
humans can handle or would take too long to do well.
So, they score well in terms of ‘game’ alignment,
integration with human skill, and
appropriateness/proof of AI model.
IIId. Core science
Hitting a flag at the top of the slalom course leads to
an increasingly difficult run lower down the slope.
The knowledge tree will be permanently misdirected
if the core measurements of the biological entities
involved in disease are erroneous. By contrast,
advances in the quality of resolution have
underpinned major breakthroughs in science, such as
Rosalind Franklin’s work leading to the discovery of
the double helix structure of DNA.
Deep Mind, working with Gladstone's Institute4, used
deep learning to enable labelling of cellular features
without having to use fluorescence tagging. The latter
technique can be constrained by the effects of spectral
overlap, inconsistency due to reagents used and even
damage to cells over time because of protocols used.
The AI system was able to match and surpass
fluorescence through using light emission z-stacks to
proxy and then become predictive of which
fluorescence was labelling. This partnership has been
able to use deep learning to gain greater visibility into
brain cells (axon or dendrite, dead or alive) which
provides important additional information when
measuring neuroplasticity or studying conditions such
as Alzheimer’s disease.
Moving downstream, once the fundamentals of
measurement and visualisation have been achieved
and we have discovered the right targets for curing
disease, we then need to be able to develop ways to
bind and affect those targets in a highly precise way.
A number of organisations are using Generative
Adversarial Networks to create a much more diverse
and high-quality library of candidate molecules than
the traditional world of drug discovery has achieved5.
This combines two different types of AI process. On
the one hand, there is a generator of many different
types of molecules that have supposedly got the right
properties to hit a target in the right way and have the
desired effects in the human body. On the other side
there is a discriminator, whose job it is to weed out
any candidates that don’t actually engage the target in
the right way. By sequentially training both to
outperform the other, the best version is finally found
and the intended result is a much-improved set of
candidates moving into human testing
33. 35
APPROACHES OF ARTIFICIAL
INTELLIGENCE IN BIOMEDICAL IMAGE PROCESSING
A Leading Tool Between Computer Vision & Biological Vision
I. INTRODUCTION
Making natural or artificial systems intelligent by
understanding the principles of computational
intelligence is the main idea of AI. It is the study
and design of an intelligent system that itself uses
its tools and techniques and widens the chances of
success. In terms of biomedical imaging, artificial
intelligence develops and implements algorithms
and strategies based on geometrical, statistical,
physical, functional etc. models and then by using
image datasets, it solves many types of problems
like visualization, feature extraction, segmentation,
image-guided surgery, texture, shape and motion
measurements, computational anatomy (i.e.
modelling
normal anatomy and its variations), computational
physiology (i.e. modelling organs and living systems
for image analysis, simulation and training),
telemedicine with medical images, etc. Due to
increased growth of medical data volume on a daily
basis, human mistakes in their manual analysis has
also been increased which in turn demands to analyze
them automatically. Therefore, usage of Artificial
Intelligence (AI) techniques in medicine proves
helpful here as it can store data, retrieves data and
provides most desirable use of information analysis
for decision making in solving problems. In the
healthcare system, treatment and diagnosis of disease
is so important in medical imaging, that for such
complex issues, algorithms of automatic medical
image analysis are helpful in providing better and
accurate understanding of medical images as well as
their increasing reliability. Therefore using intelligent
methods, accurate analysis and precise identification
of biological features can be done[3]. Such AI
methods include digital image processing and
visualization and analysis of medical images in
combination with methods like machine learning,
fuzzy logic and pattern recognition.
I. MEDICAL IMAGE SEGMENTATION (or
CLASSIFICATION)
The main purpose of image segmentation is the
division of an image into disjoint parts having a strong
correlation with objects or areas of the real world. It
is one of the most important steps in analysis of digital
images. In diagnostic and teaching
purposes in medicine, medical image classification
plays an important role. Classification basically refers
to assigning a physical object into one of a set of
predefined categories. Image classification or
segmentation can be done in many ways based on both
grey-scale and colour image analysis; Textural
analysis; Data Mining Techniques; Neural Network
Classification etc.
By medical image analysis, information like volume
measurement, description of anatomy structures by
Reliable quantitative analysis of medical images is
obtained. In later steps, other segmentation processes
like feature extraction, image measurement and
Region of interest representation are focused[8]. Then
segmentation of Region of interest is correctly carried
out by diagnosing features of disease or subsequent
lesion in medical image analysis. But unfortunately
this manual segmentation is too time-consuming and
segmentation of many scans is not possible. Therefore,
this makes intelligent tools so essential because using
them segmentation is done automatically. Various
artificial intelligence techniques such as artificial
neural networks and fuzzy logic are used for
classification problems in the area of medical
diagnosis.
34. 36
The Current Available Models For Image
Segmentation Are:-
A. Image Segmentation Using Fuzzy Logic
Fuzzy Logic, initiated in 1965 by Prof.Lotfi A.
Zadeh. is an organized method that deals with
imprecise data. Methodology- There are two
different approaches of Fuzzy logic: Region-based
segmentation, that is, classification by
thresholding, in which sets of attributes, region
growing, division and merging are being looked at.
The other is Contour based segmentation, which is,
looking for local discontinuities like derivatives
operators, mathematical morphology, etc. These
two approaches solve a dual problem, representing
each region by its closed boundary, where each
closed boundary describes a region. A two-
dimensional fuzzy image representing a real
function is taken on each pixel coordinate having
properties like brightness, texture, edginess are
defined by membership function. The aim of
contour based segmentation here is to detect fuzzy
contour of objects that represent their mechanism
and not the shape of contours.On the other hand,
the aim of region based segmentation is to use
partitional clustering , region growing and data
clustering(in hierarchical order.
B. Image Segmentation Using Artificial Neural
Networks Artificial Neural Networks are nonlinear,
nonparametric, and adaptive. Artificial Neural
Networks have successfully covered a wide variety of
real world classification such as speech recognition,
fault detection, medical diagnosis etc. Methodology-
Using arbitrary accuracy, they can theoretically
approximate any fundamental relationship. Artificial
Neural Network as a classifier is popular because it
uses iterative training by which weights representing
the solution are found. Its physical implementation
structure is simple and complex class distributions
can be easily mapped through it. Neural networks,
uses supervised and unsupervised classification
techniques. Also with the usage of ‘Self Organizing
Maps’ of neural networks, cluster based medical
image classification is done which is helpful in
‘Computer Aided Diagnostic’ decision making as
well as in categorization. The supervised
classification in Artificial Neural Network can be
achieved by using various methods like Bayesian
Decision Theory, Linear Discriminant Analysis,
Support Vector Machine; each of them offering their
unique techniques to carry out classification.
Normally the data is divided into training and testing
subsets performing classification of the image and
validating the result.
C. Image Segmentation Using Textural Classification
Applications covering Textural Classification can be:
Industrial and Biomedical Surface Inspection (for
example finding the defects and disease), ground
classification and segmentation of satellite or aerial
imagery, etc.
Methodology- In texture classification, analysis is
done based on texture of image by
subcategorization it into four methods, viz.
statistical, geometrical, and model-based and signal
processing. The process takes place as an unknown
sample image is assigned to one of a set of known
texture classes where a successful classification or
segmentation requires an efficient description of the
image texture. This technique being gigantically
adaptable that it can be applied to virtually any
modality of digital image.
Also, another concept of Wavelet transform is an
important part of textural classification which
enables evaluation of spatial frequencies at multiple
scales by designing the wavelet functions. For this
spatial information of the image,
‘Markov Random Fields’ are very popular for their
modelling of images. These models assume that the
intensity at each pixel in the image depends on the
intensities of only the neighbouring pixels.
The next process used in texture classification is
Feature Selection and Feature Extraction
techniques. The selection of features is a key factor
required for a particular data set because the
machine learning algorithm performs based on this
statistical feature. The good the quality of feature,
the best is the result obtained. Similarly, extraction
of a unique feature enables better classification
performance. It includes computing of intensity
histogram features like Mean, Standard deviation,
Energy, Entropy, Homogeneity. It is widely used in
applications like face detection, face recognition etc.
by using image processing techniques.
In Health Care Industry
D. Cloud and Medical Image Processing
Cloud computing has evolved a new computing
model that has promising characteristics to help the
healthcare industry. Medical Image Computing is an
interconnecting field of disciplines like computer
science, data science, electrical engineering,
physics, mathematics and medicine. For solving
problems related to medical images, extraction of
information from the medical images that is
clinically authentic and relevant is done and then
various computational and mathematical methods
are developed.Planning Surgery Of Brain And Skull
Base- An improved understanding of the
relationship among the lesion, adjacent critical
structures, and possible approaches of surgical
procedure is provided to the surgeon by registration
procedure, that is, combining images of Magnetic
Resonance Imaging and Computed Tomography of
the head which results in quicker operations with
less time and also provides better positioning of
craniotomies as well as reduced craniotomy size.
Localizing Electrodes In The Brain- Implantation of
electrodes over the surface of the brain helps in
locating diseases like epilepsy and it becomes easier
to operate them. Other diseases like Parkinson’s
disease can be traced by implanting the electrodes
in the subthalamic nucleus in patients to alleviate
tumors.
RELATED WORK
A lot of research work is going on in the field of image
segmentation using AI tools and techniques from
many years now. The detailed description of the
findings and work done using various AI methods are
35. 37
summarized as follows:
In 2001, High Degree B-Spline Interpolation
technique was used which improved the quality of
images in medical images and served many benefits.
Then in 2004, the technique called Wavelet Transform
& Inverse Transform which served the purpose of
Electrocardiography signal processing was used in
clinical diagnosis. By 2007, Cellular Automata
Algorithms were used which determined hypothesis
spots of breast cancer leading to diagnosis of breast
cancer.
In 2008, Denoising and Contour Extraction Algorithm
was used that led to medical image analysis, image
enhancement, image smoothing, feature extraction
and image reconstruction. In the same year, a
Distributed System for medical request processing
was implemented that helped physicians to have
diagnostic requests remotely. Temporal Recursive
Self Adaptive Filter as well as Shape –Preserved
Fitting was implemented for the pre-processing of the
medical images and X-Ray images improving the
quality of scanned images , later in the same year.
In 2009, Interactive Image Processing technique was
used which was user friendly assisting diagnostics
and was clinically useful. To serve the purpose of
medical image registration, in the same year, a Mixed
Type Registration Approach was developed that
increased the speed of registration and provided
accuracy. In the same year again, Wavelet Edge
Detection & Segmentation was found for quantitative
coronary analysis which helped in diagnosis of heart
ailments.
In the year 2010, Embedded 3D Medical Image
Processing was developed that was used in
tomographic imaging and visualization.
In 2011, Cloud Service for BL Sharing was developed
which due to its consistency, interoperability and
security provided
sharing access to the applications of BL medical
image processing.
One most remarkable work of this approach is the
very renowned Prof. Stephen Hawking, who is the
living example of using an artificial intelligence
system.
CONCLUSION & FUTURE SCOPE Result: To
conclude, we would say that this journal paper
focuses on Artificial Intelligence & Its Approaches
in Biomedical Image Processing and insights on the
working and understanding of the concepts of AI
and how a medical image is segmented using
several models. The models presented are
introduced, described and what methods of
classification are used by them is also presented
along with the better understanding of their
methodology. Each model is also represented in
terms of an example figure for their proper
understanding. And the conclusion is driven. These
models are not defined in-depth as the paper
concerns on literature review and not on deep
defining of models. Apart from the above
mentioned models, there are several other methods
like Gaussian Filters & Gabor Filter in Artificial
Neural Network Analysis; Clustering Techniques in
Data Mining etc. for image segmentation.
Radiologists can easily diagnose cancers, heart
disease, tumors and musculoskeletal disorders more
accurately by using special AI techniques in medical
imaging analysis tools.
Future Vision: Recent advances in the techniques
of AI mentioned in the paper like image processing,
machine learning, fuzzy logic, neural networking
has driven better enhancement of diagnosis
information by computer.
Image segmentation and classification algorithms
have achieved and are expected to keep gaining
these features like robustness, repeatability, least
dependent on operator, reliability and accuracy etc.
Many brain inspired projects like Blue Brain,
Google Brain, aHuman Project are some of the
specialized on-going projects in the artificial
intelligence field. Merging science and mathematics
together has always led to innovative advancements
in the medical industry and the most notable
advancements using AI technology are seen in
biomedical research and medicines which have
raised the hopes of society at another level. It has
much more to offer in the coming years provided
required support and sufficient funding is made
available.
36. 38
REFERENCES
M. Minsky Steps toward artificial intelligence Proc IRE, 49 (1) (1961), pp. 8-30
CrossRefView Record in ScopusGoogle Scholar
International Conference on VLSI, Communication & Instrumentation(ICVCI) 2011
Proceedings published by International Journal of Computer Applications® (IJCA) – A
Review Of Medical Image Classification Technique
K.H. Yu, A.L. Beam, I.S. Kohane Artificial intelligence in healthcare Nat Biomed Eng,
2 (10) (2018), pp. 719-731 CrossRefView Record in ScopusGoogle Scholar
P. Mamoshina, A. Vieira, E. Putin, A. Zhavoronkov Applications of deep learning in
biomedicine Mol Pharm, 13 (5) (2016), pp. 1445-145 CrossRefView Record in
ScopusGoogle Scholar
Y. Peng, Y. Zhang, L. Wang Artificial intelligence in biomedical engineering and
informatics: an introduction and review Artif Intell Med, 48 (2–3) (2010), pp. 71-73
Proceedings Of The International MultiConference Of Engineers And Computer Scientists
2011 Vol I, IMECS 2011, March 16-18, 2011, Hong Kong - Application of AI Techniques
in Medical Image Segmentation and Novel Categorization of Available Methods and Tools
G Anil Kumar, Int.J.Computer Technology & Applications,Vol 5 (3),851- 860 IJCTA | May-
June 2014 ISSN:2229-6093 www.ijcta.com - Analysis of Medical Image Processing and its
Applications in Healthcare Industry
IOSR Journal of Computer Engineering (IOSR-JCE) e-ISSN: 2278-0661, p- ISSN: 2278-
8727Volume 15, Issue 3 (Nov. - Dec. 2013), PP 71-78 www.iosrjournals.org - A Survey of
Image Segmentation based on Artificial Intelligence and Evolutionary Approach
Indian Journal of Science and Technology Vol. 4 No. 11 (Nov 2011) ISSN: 0974- 6846 - A
survey on artificial intelligence approaches for medical image classification By S.N. Deepa
and B. Aruna Devi
D.T.H. Lai, R.K. Begg, M. Palaniswami Computational intelligence in gait research: a
perspective on current applications and future challenges IEEE Trans Inf Technol Biomed,
13 (5) (2009), pp. 687-702
L.C. Chin, S.N. Basah, S. Yaacob, Y.E. Juan, A.K.A.B. Kadir Camera systems in human
motion analysis for biomedical applications Proceedings of International Conference on
Mathematics, Engineering and Industrial Applications 2014; 2014 May 28–30; Penang,
Malaysia (2014) p. 090006
Jiang F, Jiang Y, Zhi H, et al. Artificial intelligence in healthcare: past, present and future.
37. 39
Stroke and Vascular Neurology 2017
Patel VL, Shortliffe EH, Stefanelli M, et al. The coming of age of artificial intelligence in
medicine. Artif Intell Med 2009
Ravi D, Wong C, Deligianni F, et al. Deep Learning for Health Informatics. IEEE J Biomed
Health Inform 2017;21:4–21.
Neill DB. Using artificial intelligence to improve hospital inpatient care. IEEE Intell Syst
2013;28:92–5.
Kayyali B, Knott D, Kuiken SV. The big-data revolution in US health care: accelerating
value and innovation. 2013 http://www.mckinsey.com/industries/healthcare-systems-and-
services/ourinsights/the-big-data-revolution-in-us-health-care (accessed 1 Jun 2017).