An Introduction To Artificial Intelligence And Its Applications In Biomedical Engineering Medicine
1. An Introduction to Artificial Intelligence and Its Applications in Biomedical
Engineering & Medicine
Akash Gupta Mohit Kandpal Varun Gupta Chaman Lal Gavendra Singh
National Institute of Technology, Jalandhar,
Abstract̶ Artificial intelligence (AI) is the intelligence of
machines and the branch of computer science that aims to
create it. The basic technique in AI is to encode a problem
as a state space in which solutions are object states in that
space. Thus, problem solving can be viewed as state space
search. To search large, combinatorial state spaces,
knowledge (e.g. heuristics) and planning are required. The
use of Artificial Intelligence techniques is believed to be
the solution to current biomedical engineering problem.
This presentation addresses the marriage between two
important scientific disciplines namely, artificial
intelligence (AI) and signal processing. Signal processing
parameters are obtained through chaotic summary
measures provided by a separate agent in the model.
Although the application is illustrated with one-
dimensional signal analysis data, possible extensions can
be made to two-dimensional signals, including images. We
define AI as the science of building computational models
that emulate intelligent problem solving behaviour.
Index Terms
Í - Artificial intelligence, intelligence of
machines, biomedical engineering, signal processing, two-
dimensional signals, computational models.
1. Introduction
The field of Artificial Intelligence (AI) is connected
with modeling human intelligence and with solving
complex problems not solvable by simple or analytic
procedures.AI is concerned with issues of knowledge
representation, symbol manipulation, logic and
inference; signal processing on the other hand is
concerned with the development of numeric algorithms
based on mathematical models. Signal processing as the
science of manipulating numeric signals or data to make
explicit information that is contained within the signal.
Biomedical engineering is a discipline that advances
knowledge in engineering, biology and medicine, and
integrates human health using cross-disciplinary
activities that integrate the engineering sciences with
the biomedical sciences and clinical practice. Signal and
imaging investigations are currently a basic component
of the diagnostic, prognostic and follow-up processes.
Current advances in diagnostic examination
technologies and development of the different
modalities have made it possible to obtain high
resolution images and signals that are able to provide
more and more precise information regarding body
structure and function. These achievements have
provided clinicians with the possibility to make more
accurate and efficient diagnoses, often in a non-invasive
way. It is no coincidence that, during the last decades,
the development of automatic or semi-automatic
processing methods has impressed a big deal of interest
and effort in the areas of medical imaging, diagnostic
radiology and electrocardiography [4], in some cases
reaching the level of a practical clinical approach. The
main object is to provide a second opinion or a second
reader that can assist clinicians by integrating the
accuracy and consistency of signal and image based
diagnoses. Diagnostic data interpretation could be
considered as one of the novel steps in the routine
procedures followed by clinical professionals when
providing personalised medical regimens. The quantity
and complexity of data and information to be analysed
and managed for this task is one of the reasons behind
the renaissance of clinical decision support systems.
2. Patient Realignment in MRI
This paper describes an approach for carrying out
automatic patient realignment using a completely
separate knowledge based system which calls on a
series of low level image/signal analysis tools to extract
low level features from the images. These features are
differentiated using the knowledge based expert system
to obtain and then correct the positional inaccuracies.
Higher level features can then be extracted and used to
'fine tune ' the realignment. While the feature extraction
and image analysis operations are implemented in a
standard numerical programming language, ' C ', the
symbolic reasoning is carried out using an expert
system shell. The presentation will discuss some of the
reasons for adopting this type of approach and describe
how breaking the problem up to an intelligent overseer
driving a series of low level image processing
algorithms, has led and guided the work. The magnetic
resonance images, produced by the Picker International
VISTA 2055 .5T machine at the National Hospital for
Nervous Diseases London have, for the purposes of this
trial, a resolution of 256 x 256 pixels quantised to 12
bits (this is scaled to 8 bits for processing). A set of
features comprising simple transformations of the pixel
data are derived from the images using low level image
processing tools. Examples of such features are
perimeter, area, and central moments. Feedback exists
at two levels within the overall system. Low level
feedback which takes place within the numerical
algorithms to adjust parameters where necessary and
reject obviously incorrect results. High level feedback
which takes place between the image processing
algorithms and the expert system shell, where the shell
directs and executes the image processing algorithms to
guide matching. From a patient's initial exposure to the
2. MRI scanner, stacks of slices in sagittal, transverse and
coronal planes are obtained. These form the master set
of slices from which features are extracted. The
knowledge base is then constructed using parameters
derived from these features. One object of the expert
system was to assist in slice identification. When a
patient is rescanned at some time after the master set
was taken, it is important that slices from the most
recent set can be related to corresponding slices in the
master set. For this slice identification task it was
decided to investigate the outer skull boundary in the
transverse plane, and the brain in the transverse plane,
and determine low level features. These took the form
of physical parameters such as perimeter area central
moments, and mean grey level. For the purposes of this
trial, a set of five consecutive transverse slices were
used.
3. Major Goal Of AI
For instance, a major goal of AI is construction of an
intelligent robot, capable of perceiving, acting,
comprehending, reasoning, and learning in complex
environments. The AI field consists of six related areas:
ï· Problem solving & search
ï· Knowledge Representation
ï· Natural Language Processing (NLP)
ï· Reasoning Systems
ï· Vision & Perception
4. Applications Of Artificial Intelligence
a) Meta Neural Networks as Intelligent Agents
for Diagnosis
Success of neural networks in medical diagnosis
depends not only on learning mechanisms and network
structure but also on data quality. If insufficient data are
available, other information must be included. In this
work, expert-derived Meta knowledge supplements a
hierarchy of neural networks that together act as an
intelligent agent. Neural networks have been shown to
provide useful models for medical diagnostic problems
that require classification into two or more categories.
Both supervised and unsupervised learning approaches
have been used to successfully model biomedical data
[4]. The sensitivity, specificity, and accuracy of these
models depend not only on the network structure but
also on the training data. For some applications,
sufficient cases may not be available to reliably train the
network for some useful parameters. In these cases,
data-derived information can be supplemented with
expert supplied information through the use of meta
knowledge. In previous work, the authors have
effectively used this technique to focus attention to sub-
networks that are more limited in scope but for which
sufficient training cases are available. In the work
described here, this method is extended to include the
hierarchy of meta neural networks as one of a number
of intelligent agents that together make up a
comprehensive decision support system. The combined
system is illustrated in an application that includes
complex data such as signal processing. The overall
structure of the decision support system is shown. It
consists of three components: collection of agents, task
manager, and communicator.
ï· Agents: The agents represent the independent
decision entities and include the client, or end-
user. The client in the case of a medical
decision support system is the human decision
maker. The other agents may be any of a
variety of decision-making methodologies,
including knowledge-based systems, neural
network models, Bayesian systems, statistical
results, and signal processing algorithms. Each
agent utilizes its own sources of information,
which may be shared with other agents. For a
patient analysis problem, domain information
is used in conjunction with patient-specific
information to arrive at the final decision.
ï· Task Manager: The principal components of
the control structure are the task manager and
the communications interface. The task
manager breaks the problem into subtasks that
are then directed toward the appropriate
agents. It also combines results from agents,
including the client, for the overall response to
the problem
.
ï· Communicator: The communicator must
present input to each agent in a form that it can
understand as well as interpret output from
each agent so that other relevant agents can
understand it. This process is illustrated in Fig.
2. The three main components are the
symbolic to numeric converter, the numeric to
symbolic converter, and the common language
generator that interprets output in a form that
can be communicated to each agent. The
objective is to overlay the conversion similarly
to a human user interface so no agent
modification is required.
3. Figure.1 Decision support system
Figure.2 Communicator
b) Medical Imaging
The Medical Imaging group works with medical image
of different modalities e.g. CT, MRI and CCD. The data
sets are real patient radiological images and the Visible
Human Project Data Set (VHP). The images of different
modalities are segmented and augmented with the
images. The work that has been carried out so far, can
be categorized as:
1. Computer Aided Design model (CAD) of femur
implants.
2. Facial Reconstruction
3. Virtual Heart
4. CBIR Database of blood filing
5. Orthodontic Treatment Simulation.
6. Database of anthropometric
7. Registration and Image
A. Guidance System
Besides, utilizing the images for making better implants
and reconstruction, considerable effort was focused
towards visualization and to the computational aspect of
the problem. In the area of visualization, effort was
directed towards development of software in three
dimensions and to incorporate virtual reality effect. The
computational aspect utilizes high end symmetric
processing like Silicon Graphics and also in the use of
clusters. The research in Medical Imaging can be
classified into the following areas:
1. CAD model of human implants and prototyping.
2. Virtual three-dimensional modeling for
reconstruction.
3. Virtual model of the human heart.
4. Finite Element Analysis (FEA) (structural analysis)
of bone, heart, implants, etc.
5. Investigation on the usage of High Performance
Computing (HPC) in the area of medical imaging.
B. A patient's prognosis
A patient's prognosis can be affected by many factors,
such as the type of cancer the patient has, the stage of
the cancer, or its grade (how aggressive the cancer is or
how closely the cancer resembles normal tissue). Other
factors that may also affect a person's prognosis include
the patient's age and general health or the effectiveness
of treatment. Statistical analysis of survival data is
usually carried out to help estimate prognosis. Survival
statistics indicate a cohort of patients with certain types
and stages of cancer and the outcome and survival
following treatment. However, statistics alone may not
be sufficient to predict the outcome of a particular
patient, as no two patients are exactly alike. These
methods are usually used to explain the data and to
model the progression of the disease rather than to make
survival predictions for populations or individual
patients. An alternative to the statistical analysis of
survival data is in using artificial neural network
technology. Neural network has not been tested
extensively for modelling survival data, but based on its
predictive success in other domains; it is considered a
good alternative for the prediction of survival of
individual patients.
c) Hybrid Intelligence for Bio-Medical
Informatics
Hybrid intelligent systems are becoming popular due to
their capabilities in handling many real world complex
problems, involving imprecision, uncertainty and
vagueness, high-dimensionality. They provide us with
the opportunity to use both, our knowledge and row
data to solve problems in a more interesting and
promising way. This multidisciplinary research field is
in continuous expansion in the artificial intelligence
research community. It is commonly accepted that there
are two parts to health sciences, the study, research, and
knowledge of health and the application of that
knowledge to improve health, cure diseases, and
understand how humans function. This configuration
(theory elicitation and theory application) is analogous
with the know-how managed by physicians that apply a
4. mixture of objective knowledge and subjective
knowledge.
d) Decision support in heart failure through
processing of electro- and Echocardiograms
HF is a progressive disorder caused by a decreased
ability of the ventricle to fill with or eject blood and in
which damage to the heart causes weakening of the
cardiovascular system. It usually manifests itself via
fluid congestion or inadequate blood flow to tissues and
progresses to underlying heart injury or inappropriate
responses of the body to heart impairment.
Unfortunately, HF is a progressive disorder that must be
managed with regard, not only to the state of the heart,
but also to the condition of the circulation, lungs,
neuroendocrine system and other organs as well. In its
chronic form, HF is one of the most remarkable health
problems in terms of prevalence and morbidity,
especially in the developed western countries, with a
strong impact in terms of social and economic effects.
All these aspects are typically emphasised within the
elderly population, with very frequent hospital
admissions and a significant increase of medical costs.
With the aim of avoiding unnecessary generality, this
paper addresses the specific, yet complex and
paradigmatic example of image and signal processing
for decision support in heart failure (HF). HF is a
clinical syndrome, whose management requires from
the basic diagnostic workup the involvement of several
stakeholders and the exploitation of various imaging
and nonimaging diagnostic resources. Indeed, given the
complexity of the management of chronic HF patients,
several attempts to address the problem have been made
in various research projects and have resulted in the
development of dedicated information technology
solutions, such as automated guidelines systems,
decision support systems, or machine learning methods
for automated HF diagnosis or prognosis. The
HEARTFAIL platform has been conceived as an
integrated and interoperable system, able to guarantee
an umbrella of services ranging from the acquisition
and management of raw data to the provision of
effective decisional support to clinicians. Specifically,
the core of the platform is represented by a CDSS,
which has been carefully designed by combining
innovative knowledge representation formalisms, robust
and reliable reasoning approaches, innovative methods
for diagnostic image analysis, and robust and high-
performance algorithms for signal processing.
Figure.3 HF diagnostic workflow
On the other hand, the ECG is recognised as the most
fundamental examination performed in the evaluation
and assessment of several heart abnormalities, including
HF. A typical ECG tracing of a cardiac cycle consists of
a P wave, a QRS complex (structure on the ECG that
corresponds to the depolarisation of the ventricle and is
composed of the Q, R and S waves) and a T wave.
According to [23], the negative predictive value of a
normal ECG in excluding left ventricular systolic
dysfunction exceeds 90%. The most common ECG
examinations are the resting ECG and the Holter ECG.
While the latter is more commonly used for the
discovery of rhythm abnormalities and the computation
of heart rate variability, the former is more commonly
used for the evaluation of morphological abnormalities
in the heartbeat. Decision support in HF Recent studies
and experience have demonstrated that accurate heart
failure management programs, based on a suitable
integration of inpatient and outpatient clinical
procedures, might prevent and reduce hospital
admissions, improving clinical status and reducing
costs. Routine practice in HF cases presents several
aspects on which automatic, computer-based support
could have a favourable impact. A careful investigation
of the needs of HF practitioners and the effective
benefits assured by decision support was performed,
and four problems were identified as highly beneficial
for CDSS point-of-care intervention. They can be
described as macro-domain problems and listed as: (i)
HF diagnosis, (ii) prognosis, (iii) therapy planning, and
(iv) follow up. Further detailed decision problems have
been identified for specifying these macro-domains,
focusing as much as possible on the medical usersâ
needs; indicative examples are:
ï· Evaluation of HF severity;
ï· Identification of suitable pathways;
ï· Planning of adequate, patient-specific therapy;
ï· Analysis of diagnostic examinations;
ï· Early detection of patientâs decompensation.
The idea behind the development of a CDSS able to
support this kind of problem has been to provide
clinicians with advice, suggestions and alerts in the
different phases of management of chronic HF patients,
without altering their normal activities.
5. e) Detecting wrong blood in tubeâ errors:
Evaluation of a Bayesian network approach
Medical errors are a significant problem in the United
States. They kill more Americans each year than motor
vehicle accidents, breast cancer.In laboratory medicine,
of particular concern are patient identification errors.
Proper patient identification is essential to reducing
errors and improving patient safety.
f) Brain Computer Interface
A BCI system is a communication system in which
messages or commands that an individual sends to the
external world do not pass through the brainâs normal
output pathways of peripheral nerves and muscles but is
detected through EEG activity. Nowadays BCI has become
a popular research topic in the biomedical signal
processing area. In the analysis of electromagnetic (EM)
brain signals it is required to extract neuro physiologically
meaningful information either for clinical reasons, such as
is the case with the analysis of the epileptic
electroencephalogram (EEG) to extract information on
underlying epileptogenic sources. Similarly EEG can be
used as in the field of Brain-Computer Interfacing (BCI)
where brain signals are interpreted to provide a means of
communication.
5. Discussion
The structures of the agents themselves include both
expert-derived and data-derived domain knowledge,
allowing the inclusion of all available data. The
hierarchical structure of the neural network addresses
the serious problem of missing data and allows focus on
smaller networks that can be more easily and accurately
trained on smaller data sets.
6. Conclusion
In this paper we go through different research papers
whose references are listed and after that we are able to
conclude that with the help of artificial intelligence,
biomedical engineering has been taken new direction
towards the success ,reliable ,easiness and accurate.
Accuracy is very important in biomedical clinical
applications. In laboratory medicine, of particular
concern are patient identification errors. Proper patient
identification is essential to reducing errors and
improving patient safety. Signal and image processing
methods may be understood and embedded as a part of
a model base of the CDSS. In such a way, it is possible
to achieve an effective high-level integration of signal
and image processing methods into the general process
of care. In this paper a CDSS for the management of HF
has been presented. The wide range of services
provided by the CDSS is enabled by a high-level
integration of diagnostic signal and image processing.
REFERENCES
[1]. D. B. Shannanl and T. S. Durrani âAn Overview of
AI Applied to Signal Processing: A Perspective on
Coupled Systemsâ,1989.
[2]. Nagao M., Matsuyama T., Mori H.,"Structural
Analysis of Complex Aerial Photographs",UCAI-
79, Tokyo, August 79.
[3]. Miiios E.,"Interpretation-Guided Signal Processing
via Protocol Analysis", Proc IEEE ICASSP 85,
Tampa, Vol. 4, pp1660-1663.
[4]. Dellepiane S., Serpico S. B., Venzano L., Vemazza
G.,"Structural Analysis in Medical
Images",ESACONTROL, Biomedical Division,
1985.
[5]. Maurice E. Cohen, Donna L. Hudson âMeta
Neural Networks as Intelligent Agents for
Diagnosisâ, 2002 IEEE.
[6]. C. J. James âBlind Source Separation in single-
channel EEG analysis: An application to
BCIâ,IEEE ,China 2001.
[7]. C.J. James and D. Lowe, âExtracting multisource
brain activity from a single electromagnetic
channel,â Artificial Intelligence in Medicine, vol.
28, issue. 1, pp. 89â104, 2003.
[8]. C.J. James, M. Davies and O. Gibson, âOn the
analysis of Single versus Multiple Channels in
Electromagnetic Brain Signal Analysis within a
Dynamical Embedding Framework,â Artificial
Intelligence in Medicine, to appear, 2006.
[9]. C.J. James and C.W. Hesse, âA comparison of time
structure and statistically based BSS methods in the
context of long-term epileptiform EEG recordingsâ,
Proc. Int. Conf. on Independent Component
Analysis and Blind Signal Separation (ICA2004),
Granada, Spain, 2004.
[10]. J. Vidal, âToward direct brain-computer
communicationâ Annu. Rev. Biophys. Bioeng. pp.
157â180, 1973.
[11]. S. Makeig, A.J. Bell, T.-P. Jung, and T.-J.
Sejnowski., Independent component analysis of
electroencephalographic data., In: Advances in
Neural Information Processing Systems 8, MIT
Press 145-151, 1996.
[12]. C.J. James and C.W. Hesse, âIndependent
Component Analysis for Biomedical Signals,â
Physiological Measurement 26 R15âR39,2005.
[13]. A.J. Bell and T.J. Sejnowski, âLearning higher
order structure of a natural sound,â Network, 7 261-
266, 1996.
[14]. C.J. James and C.W. Hesse, âMapping Scalp
Topographies of Rhythmic EEG Activity using
Temporal Decorrelation based Constrained ICA,â
Proc. of 27th Annual International Conference of
the IEEE Engineering in Medicine and Biology
Society, Shanghai, China, 4 pages (CDROM),
2005.
[15]. B. Blankertz, K.-R. MĂŒller, T. V. G. Curio, G.
Schalk, J. Wolpaw, A. Schlögl, C. Neuper, G.
Pfurtscheller, T. Hinterberger, and M. S. N.
Birbaumer, âThe BCI competition 2003: Progress
and perspectives in detection and discrimination of
6. EEG single trials,â IEEE Trans. Biomed. Eng., vol.
51, pp. 1044â1051, 2004
[16]. G. Pfurtscheller and A. Arabibar, âEvaluation
of event-related desynchronization preceding and
following voluntary self-paced movement,â
Electroenceph. Clin. Neurophysiol., vol. 46, pp.
138â46, 1979.
[17]. Woo Chaw Seng et al., âImage Segmentation
of MRI images using wavelet transform and
artificial neural networkâ, proceedings of the 7th
JSPSVCC Seminar in Integrated Engineering,
University of Malaya, 1998, pp.32-37.
[18]. Mangalam S, N. Selvanathan, Sapiyan Baba,
âMaterial Classification of 3D voxel of MRI using
Bayesianâ , 2000,National Conf. On Biomedical
Engineering, Kuala Lumpur.
[19]. N. Selvanathan, S. Somasundaran, David
Choon, John George, âDesign of medical implants
from Magnetic Resonance (MR) scanned images
for the human femurâ IASTED International
Conference on Computer Graphics and
Imaging,2001, Hawaii, USA.
[20]. Sameem Abdul Kareem, Sapiyan Baba, Mohd
Ibrahim A Wahid, âResearch In Medical
Informaticsâ. Health Informatics Journal, Issue 6.2,
Sheffield Academic Press Ltd., Sheffield England.
June 2000.
[21]. Sameem Abdul Kareem, Sapiyan Baba, Yong
Zulina Zuhairi, Mohd Ibrahim A Wahid, âANN As
A Tool For Medical Prognosis Health Informatics
Journal, Issue 6.3, Sheffield Academic Press Ltd.,
Sheffield England. September 2000.
[22]. Emilio Corchado âSpecial Issue: Hybrid
Intelligence for Bio-Medical
Informaticsâ,IEEE1994, Iran.
[23]. Franco Chiarugi Decision support in heart
failure through processing of electro- and
echocardiograms Artificial Intelligence in Medicine
50 (2010) 95â104.
[24]. Rihal CS, Davis KB, Kennedy JW, Gersh BJ.
The utility of clinical, electrocardiographic, and
roentgenographic variables in the prediction of left
ventricular function. Am J Cardiol 1995;75(4):220â
3.
[25]. Tierney WM, Overhage JM, Murray MD,
Harris LE, Zhou XH, Eckert GJ, et al. Effects of
computerized guidelines for managing heart disease
in primary care.J Gen Intern Med 2003;18(12):967â
76.
[26]. Kim J,Washio T, YamagishiM, Yasumura Y,
Nakatani S, Hashimura K, et al. A novel data
mining approach to the identification of effective
drugs or combinations for targeted endpointâ
applications to chronic heart failure as a new form
of evidence-based medicine. Cardiovasc Drugs
Ther 2004;18 (6):483â9.
[27]. Colantonio S, Martinelli M, Moroni D, Salvetti
O, Perticone F, Sciacqua A, et al. Decision support
and image & signal analysis in heart failure. A
comprehensive use case. In: Azevedo L, Londral
AR, editors. HEALTHINF 2008. Proceedings of
the first international conference on health
informatics. Setubal, Portugal: INSTICC-Institute
for Systems and Technologies of Information,
Control and Communication; 2008. p. 288â95.