2. What is AI?
⢠Systems that think like humans
⢠Systems that act like humans
⢠Systems that think rationally
⢠Systems that act rationally
3. AI and its concepts
⢠Artificial intelligence (AI) is an area of computer science
that emphasizes the creation of intelligent machines
that work and react like humans.
⢠More on this:
https://www.datacamp.com/community/tutorials/machi
ne-deep-
learning?utm_source=adwords_ppc&utm_campaignid=1
455363063&utm_adgroupid=65083631748&utm_device
=c&utm_keyword=&utm_matchtype=b&utm_network
=g&utm_adpostion=&utm_creative=278443377092&ut
m_targetid=aud-748597547652:dsa-
429603003980&utm_loc_interest_ms=&utm_loc_physic
al_ms=9076813&gclid=Cj0KCQjwp4j6BRCRARIsAGq
4yME2pUHqSEVMYTpfihH2ju8mrpCFtrmA0vHEP2w
ZAkWcbWEvSEEYWhoaAh4tEALw_wcB
4. Definition: Artificial intelligence (AI) refers to the simulation of human
intelligence in machines that are programmed to think like humans and mimic
their actions
5. ⢠Machine learning: This is the practice of using algorithms to
explain/analyse/parse data, learn from it, and then make them apply what
theyâve learned to make informed decisions based on what it has learnt.
⢠A ML workflow starts with relevant features being manually extracted
from images/data. The features are then used to create a model that
categorizes the objects in the image.
⢠Examples: decision trees, support vector machines, neural networks,
ensemble methods and naĂŻve bayes classifier.
6. Deep Learning
⢠DL is an Artificial intelligence function that imitates the workings of the human
brain in processing data and creating patterns for use in decision making
⢠DL is a subset of machine learning in artificial intelligence (AI) that has networks
capable of learning unsupervised from data that is unstructured or unlabeled.
⢠In a deep learning workflow, relevant features are automatically extracted from
images/data. A network is given raw data and a task to perform, such as
classification, and it learns how to do this automatically.
Deep learning improves as the size of your data increases
7. ⢠Deep learning: Deep learning is an artificial intelligence function that
imitates the workings of the human brain in processing data and creating
patterns for use in decision making
⢠Deep learning AI is able to learn from data that is both unstructured and
unlabeled e.g e-mail messages, word processing documents, videos, photos, audio files,
presentations, webpage
⢠Examples of DL models: ANNS
8. DL models
Artificial Neural Networks(ANN): The artificial neural networks are built
like the human brain, with neuron nodes connected together like a web.
9. Choosing Between Machine Learning and
Deep Learning
⢠ML offers a variety of techniques and models you can choose based on your
application, the size of data you're processing, and the type of problem
you want to solve.
⢠A successful deep learning application requires a very large amount of data
(thousands/tons of images) to train the model, as well as GPUs, to rapidly
process your data.
⢠Graphics Processing Unit: A Graphics Card is a piece of computer hardware that
produces the image you see on a monitor.
10. ML in medical diagnosis
⢠Use cases make near perfect diagnosis, recommend best medicines, predict
readmissions and identify high-risk patients.
⢠Such predictions are based on the dataset of anonymized patient records and
symptoms exhibited by a patient.
11. ⢠To choose ML or DL, consider whether you have a high-performance
GPU and lots of labeled data.
⢠If you donât have either of those things, it may make more sense to use
machine learning instead of deep learning.
⢠Deep learning is generally more complex, so youâll need at least a few
thousand images to get reliable results.
⢠Having a high-performance GPU means the model will take less time to
analyze all those images.
12. Use cases for AI in the imaging world
1. Identifying cardiovascular abnormalities: Measuring the various structures of the heart
can reveal an individualâs risk for cardiovascular diseases or identify problems that
may need to be addressed through surgery or pharmacological management.
How?
Automating the detection of abnormalities in CXR e.g Thickening of certain
muscle structures, such as the left ventricle wall, changes in blood flow through the heart and
associated arteries need to be monitored
This could lead to quicker decision-making and fewer diagnostic errors
13. Use cases for artificial intelligence in the
imaging world
2. Detecting fractures and other musculoskeletal injuries:
Using AI to identify hard to see fractures, dislocations, or soft tissue injuries
could allow surgeons and specialists to be more confident in their treatment
choices.
Difficult fractures to detect: odontoid fracture(a type of fracture in the
cervical spine)
AI can be used to see undetected variations in the image that could indicate an
instability that requires surgery.
14. Use cases for artificial intelligence in the
imaging world
3. Aiding in the diagnosis neurological diseases: e.g amyotrophic lateral
sclerosis (ALS) vs primary lateral sclerosis (PLS).
Radiologists must decide if lesions are relevant or simply mimicking the
structures of one of the diseases, and false positives are relatively common.
Current method: manual segmentation and quantitative susceptibility
mapping(QSM). These are difficult, and time consuming
15. Solution: Automating this procedure with ML would facilitate research and
assist in the development of a promising imaging biomarker.
⢠Algorithms may be able to streamline this process by flagging images that
indicate suspect results and offering risk ratios that the images contain
evidence of ALS or PLS.
⢠Algorithms may also be able to automatically populate reports, reducing
workflow burdens on providers.
16. 4. Flagging thoracic complications and conditions
Diagnose pneumonia and distinguish it from other lung conditions like
bronchitis.
ď§ What if a radiologist is not available?
ď§ Radiologists may have difficulty identifying pneumonia if the patient has pre-
existing lung conditions, such as malignancies or cystic fibrosis
17. ⢠An AI algorithm could assess x-rays and other images for evidence of
opacities/opaqueness(not transparent) that indicate pneumonia, then alert
providers to the potential diagnoses to allow for speedier treatment.
5. Screening for common cancers:
Breast cancer: Micro-calcification in tissue can often be difficult to
conclusively identify as either malignant or benign.
False positives could lead to unnecessary invasive testing or treatment, while
missed malignancies could result in delayed diagnoses and worse outcomes.
18. ⢠AI can help improve accuracy and use quantitative imaging features to more
accurately categorize micro-calcifications by level of suspicion(size, shape
etc) potentially decreasing the rate of unnecessary benign biopsies
⢠Providing risk scores for areas of concern could allow providers and patients
to make more informed decisions about how to proceed with testing or
treatment.
19. ⢠Using AI in polyp detection at CT Colonography (CTC) can reduce false
positives and reduce medical legal risk for radiologists.
6. Example: Identification of vascular flow alterations during the subclinical
phase of the DR could provide timely recognition of patients at a greater risk
of DR progression
20. Applications of AI in Ultrasound
1. In vascular ultrasound, AI aids in automatically tracing spectral waveform with
high accuracy
2. In fetal echocardiography, AI aids in detection of subtle abnormal cardiac rhythms
3. In breast ultrasound, AI can be used in detecting subtle changes in breast tissue
that are not easily detectable with the human eye
4. In prostate ultrasound, AI can detect patterns that may be more consistent with
malignant versus benign changes in the prostate gland in men of advanced age
21. Applications of AI in Ultrasound
⢠Patient scheduling and preparation: prioritizing work lists and identifying high risk
patients using relevant priors/clinical information.
⢠Standardising imaging protocols and acquisition: preâprocessing tasks such as
reconstruction, registration and segmentation, quality optimisation, image navigation, multi
transducer technologies, eg Novel 3DUS imaging methodologies.
⢠Image and data interpretation: Real time biometrics and anomaly detection or offline 3D
applications, and new computer interface design for AI application plugins.
⢠Reporting and recommendations: Automated structured reporting versus narrative.
Decision support for follow up of recommendations in the case of abnormalities.
22. Applications of AI in Ultrasound
⢠Breast mass detection and diagnosis with ultrasound:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6403465/
⢠Diagnosis of Esophageal Cancer:
https://healthitanalytics.com/news/machine-learning-tool-
accurately-diagnoses-esophageal-cancer
23. Applications of AI in Ultrasound
⢠Breast Imaging challenges: lengthy exam and reading times, relatively
high number of false positives and the lack of repeatability of results due to
operator dependence.
Using AI to solve these challenges: QView Medical
https://www.intelligentultrasound.com/
AI in Obs/Gyn: https://www.contemporaryobgyn.net/obstetrics/artificial-
intelligence-obgyn-ultrasound
24. Advance Artificial Intelligence in Medical Imaging?
How?
NIH, RSNA(2018) and The Academy for Radiology and Biomedical Imaging
Research
Steps:
1. Develop standardized use cases: standard inputs and outputs among
comparable algorithms. A standard way of accepting inputs and outputs for
algorithms to process will be required e.g Extensible Markup Language, DICOM
etc
These create standards for validation before AI algorithms are ready for clinical use
25. 2. Establish data sharing methods
Need for quality datasets that contain appropriate annotations or rich
metadata. This ensures scalability of the algorithms.
3. Ensure technology is safe and accurate
Involve IT developers, government agencies, and the public to make sure AI
algorithms are accurate, free of bias, and safe for patients
This is important for validation.
26. ⢠AI in Radiology and pathology: https://healthitanalytics.com/news/how-
artificial-intelligence-is-changing-radiology-pathology
⢠Radiologist, AI Combination Improves Breast Cancer Detection
https://healthitanalytics.com/news/radiologist-ai-combination-improves-
breast-cancer-detection
27. Challenges of AI
⢠Bad data will result in bad models
âfollow good data practices in both the creation and curation of retrospective datasets for
model training and in the prospective collection of the dataâ.
Data is at the center of everyday healthcare tasks and broader industry
improvements, making it an incredibly valuable resource for organizations
28. Examples of ML methods
⢠Artificial Neural Networks: ANN is a computational model based on the
structure and functions of biological neural networks. Information that
flows through the network affects the structure of the ANN because a
neural network changes - or learns, in a sense - based on that input and
output.
⢠ANN is an implied model of the biological neuron to make decisions and
conclusions by
29. ⢠Input layer: It can be pixels of an
image or a time series data
⢠Hidden layer: Commonly known
as weights which are learned while
the neural network is trained
⢠Output layer: Final layer mainly
gives you a prediction of the input
you fed into your network.
30. output = sum (weights * inputs) + bias
⢠bias acts like a constant which allows you to shift the activation function to either
right or left thus helping the model to fit the given data.
⢠The steepness of the Sigmoid depends on the weight of the inputs.
⢠Y=mx+c
⢠It allows you to move the line down and up fitting the prediction with the data
better. If the constant c is absent then the line will pass through the origin (0, 0) and
you will get a poorer fit.
31. ⢠Depending on the output if the
output is not the desired one, go
back and adjust the weights(back
propagation) and train again.
35. ⢠Support Vector Machines(SVM)
⢠Decision trees, k-nearest neighbor, Naïve Bayes,, logistic regression.
⢠Other methods can be found: https://towardsdatascience.com/10-
machine-learning-methods-that-every-data-scientist-should-know-
3cc96e0eeee9
36. Examples of ML methods
Application of Neural Networks in Medical Image Processing
⢠http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.403.4346&rep=r
ep1&type=pdf
Medical image analysis with artificial neural networks https://sci-
hub.tw/https://www.ncbi.nlm.nih.gov/pubmed/20713305
37.
38. Supervised and Unsupervised Learning
Supervised Learning:
⢠Input data is called training data and has a known label or result such as
spam/not-spam or a stock price at a time(classification and regression
problems)
⢠A model is prepared through a training process in which it is required to
make predictions and is corrected when those predictions are wrong.
⢠The training process continues until the model achieves a desired level of
accuracy on the training data.
39. Supervised Learning
⢠Supervised learning models have a baseline understanding of what the
correct output values should be.
⢠With supervised learning, an algorithm uses a sample dataset to train itself to
make predictions, iteratively adjusting itself to minimize error. These datasets
are labeled for context, providing the desired output values to enable a model
to give a âcorrectâ answer.
40. Categories of supervised learning
⢠Classification problems use an algorithm to accurately assign test data into specific
categories, such as separating apples from oranges. Or, in the real world, supervised learning
algorithms can be used to classify spam in a separate folder from your inbox.
⢠Support vector machines, decision trees and random forest are all common types of classification
algorithms.
⢠Regression is another type of supervised learning method that uses an algorithm to
understand the relationship between dependent and independent variables. Regression
models are helpful for predicting numerical values based on different data points, such as
sales revenue projections for a given business.
⢠Some popular regression algorithms are linear regression, logistic regression and polynomial regression.
41. Examples of supervised learning algorithms
⢠Nearest Neighbor
⢠Naive Bayes
⢠Decision Trees
⢠Linear Regression
⢠Support Vector Machines (SVM)
⢠Artificial Neural Networks
⢠Random Forests
42. Supervised and Unsupervised Learning
Unsupervised Learning
⢠Input data is not labeled and does not have a known result.
⢠A model is prepared by deducing structures present in the input data. This
may be to extract general rules.
⢠It may be through a mathematical process to systematically reduce
redundancy, or it may be to organize data by similarity(Clustering problem)
⢠Example: k-means clustering
43. Unsupervised Learning
⢠Unsupervised learning algorithms work independently to learn the data's
inherent structure without any specific guidance or instruction.
⢠You simply provide unlabeled input data and let the algorithm identify any
naturally occurring patterns in the dataset.
⢠Unsupervised learning is more helpful for discovering new patterns and
relationships in raw, unlabeled data.
44. Categories of Unsupervised Learning
1. Clustering is a data mining technique for grouping unlabeled data based on their similarities or
differences. For example, K-means clustering algorithms assign similar data points into groups,
where the K value represents the size of the grouping and granularity. This technique is helpful for
market segmentation, image compression, etc.
2. Association is another type of unsupervised learning method that uses different rules to find
relationships between variables in a given dataset. These methods are frequently used for market
basket analysis and recommendation engines, along the lines of âCustomers Who Bought This Item
Also Boughtâ recommendations.
3. Dimensionality reduction is a learning technique used when the number of features (or
dimensions) in a given dataset is too high. It reduces the number of data inputs to a manageable size
while also preserving the data integrity. Often, this technique is used in the preprocessing data stage,
such as when autoencoders remove noise from visual data to improve picture quality.
45. Supervised and Unsupervised Learning
⢠Supervised machine learning is suited for classification and regression tasks,
such as weather forecasting, pricing changes, sentiment analysis, and spam
detection.
⢠While unsupervised learning is more commonly used for exploratory data
analysis and clustering tasks, such as anomaly detection, big data
visualization, or customer segmentation.
46. How to choose between supervised and
unsupervised learning
1. Is your data labeled or unlabeled? Supervised learning requires labeled datasets.
Youâll need to assess whether your organization has the time, resources, and expertise
to validate and label data.
2. What are your goals? Itâs important to consider the type of problem youâre trying
to solve and whether you are trying to create a prediction model or looking to discover
new insights or hidden patterns in data.
3. What types of algorithms do you need? When deciding what approach is best
suited for your organization, itâs also important to evaluate if there are algorithms that
can support the volume of data and match the required dimensions, such as the
number of features and attributes.
47. Other key differences between supervised and
unsupervised learning
1. Goals: In supervised learning, the goal is to predict outcomes for new data.
You know up front the type of results to expect. With an unsupervised learning
algorithm, the goal is to get insights from large volumes of new data. The
machine learning itself determines what is different or interesting from the
dataset.
2. Applications: Supervised learning models are ideal for spam detection,
sentiment analysis, weather forecasting and pricing predictions, among other
things. In contrast, unsupervised learning is a great fit for anomaly detection,
recommendation engines, customer personas and medical imaging.
48. Other key differences between supervised and
unsupervised learning
3. Complexity: Supervised learning is a simple method for machine learning, typically
calculated through the use of programs like R or Python. In unsupervised learning,
you need powerful tools for working with large amounts of unclassified data.
Unsupervised learning models are computationally complex because they need a large
training set to produce intended outcomes.
4. Drawbacks: Supervised learning models can be time-consuming to train, and the
labels for input and output variables require expertise. Meanwhile, unsupervised
learning methods can have wildly inaccurate results unless you have human
intervention to validate the output variables.
49. Semi- Supervised Learning
⢠Not sure that either of these options is the right fit? You could also consider a third
approach: semi-supervised learning.
⢠Semi-supervised learning combines aspects of both supervised learning and unsupervised
learning. Machine learning techniques that fall under this category utilize both labeled and
unlabeled data to train a predictive model.
⢠Semi-supervised learning uses a small amount of labeled data to train an initial model, which
can be used to predict labels on a larger amount of unlabeled data. The model is then
applied iteratively to both originally labeled data and data with predicted labels (pseudo-
labels). After, you will add your most accurate predictions to the labeled dataset and repeat
the process again to continue improving the performance of your model.
50. Predictive Analytics
Definition:
Predictive analytics can be described as a branch of advanced analytics that is
utilized in the making of predictions about unknown future events or activities
that lead to decisions.
51. Benefits of predictive analytics/modelling
1. Improved patient outcomes. By integrating patient records with other health data,
healthcare organizations can detect warning signs of serious medical events and proactively
prevent their occurrence.
By looking at data and outcomes of past patients, machine learning algorithms can be
programmed to provide insight into methods of treatment that will work best for the current
patients.
2. Holistic health support. Evolving patient-centric models focus on the person as a whole
rather than on outcomes in isolation. Predictive tools make it possible to collect and integrate
lifestyle, symptom and treatment data to produce holistic treatment plans.
3. Enhanced operations. Predictive tools can also be applied to internal healthcare processes
such as equipment provision or staffing requirements to help lower overall costs.
52. 4. Personalized service provision. Care personalization has taken center stage as
pandemic pressures evolve. Predictive tools make it possible to create truly
personalized treatment plans tailored to unique patient needs.
5.Aides decision Making process: Determining whether treatment is likely to work
or not. The model could assess how a patient is responding to treatment months
earlier than previously available.
This will allow providers to switch their course of treatment sooner if the current one
is not working, saving patients months of unnecessary and painful treatment. Overall,
this will improve the quality of patient care
53. 6. Relief for healthcare workers: Predictive analytics lightens the load for
healthcare works by assisting in the diagnosis process
AI algorithm to predict Alzhiemerâs disease:
https://healthitanalytics.com/news/app-uses-artificial-intelligence-algorithm-
to-predict-alzheimers
Editor's Notes
a programme which mimics human cognition
deep learning is just a subset of machine learning
The Graphics Card is responsible for rendering an image to your monitor, it does this by converting data into a signal your monitor can understand.
ML: Diagnosis, Treatment recommendation,Image recognition
odontoid fracture: Fracture in the neck that occurs as a result of trauma to the cervical spine
An imaging biomarker is a feature of an image relevant to a patient's diagnosis
RR: the risk of a health event among one group with the risk among another group.
There is variability in Radiologist interpretation
Example of a use case: screening for breast cancer
Who?
medical imaging community(radiologists,academic institutions etc)
Creating models for validation and monitoring of AI algorithms and minimizing unintended bias will require collaborations between researchers, industry developers, and government agencies.
ANNs simulate human brainâs work.
Just like biological neural network, artificial neural network is constantly learning and updating its knowledge and understanding of the environment based on experiences that it encountered
And |a| and |b| are the length of string a and b respectively with variables đŁ 1 ⌠đŁ đ
A RF consists of a large number of individual decision trees that operate as an ensemble.
Each individual tree in the random forest spits out a class prediction and the class with the most votes becomes our modelâs prediction