Fuzzy Logic in Medicine and Biology

Bill Thompson wgthom at gandalf.rutgers.edu
Tue Sep 13 20:02:58 EST 1994


Dear  Colleague:

I am enclosing the detailed part of the tutorial talks and
some of the medical applications that will be presented by Drs. Akay, 
Hudson, Cohen and Cios.

In addition to these, we will have other medical and biological
applications which will be presented by invited speakers in the
field.

Note that the workshop will be held Nov 1st and 2nd which is before 
the Conference. The fee for the workshop is

IEEE Member: $250
Others:      $330

The conference will start Nov 3 and end Nov 6. Attendees to the
workshop do not have to continue with  the conference, although I
highly recommend it. Note that the conference will require a separate
attending fee. 

Please do not hesitate to call me if you have any questions
regarding the workshop. I can be reached at (908) 445-4906
or (akay at gandalf.rutgers.edu).


I am looking forward to hearing from you.


Sincerely,

Metin AKAY, Ph.D.
Phone:  (908) 445-4906
e-mail: akay at gandalf.rutgers.edu

                WORKSHOP
     Fuzzy Logic in Medicine and Biology


First Day

Participation:     Dr. Metin Akay , Rutgers University
                   Dr. Donna Hudson, UCSF
                   Dr. Maurice Cohen, UCSF
                   Dr. Krzysztof Cios, Univ Of Toledo

Fundamentals of Fuzzy Logic


Imprecisely defined classes play an important role in human thinking.
Fuzzy set theory derives from the fact almost all natural classes and
concepts are Fuzzy rather than crisp in nature. According to
Lotfi Zadeh, who is the founder of fuzzy logic,  all of the
reasoning people use everyday is Approximate in nature. Humans
work from approximate data, extract meaningful information from massive
data, and find crisp solutions. Every moment, for example, the
photoreceptors of our retina are bombarded with millions of light
rays, and we perceive and understand the total picture as clear and
sunny day. The fuzzy logic provides a suitable basis for the ability
to summarize information, to extricate from the collections of masses
of data impinging upon the human brain those and only those which are
relevant to performance of the task at hand.

In this workshop, the concepts of fuzzy logic including the fuzzy
logic controller, fuzzy pattern recognicition systems, the fuzzy expert
system, and fuzzy neural networks for biological control systems,
pattern recognition and expert systems will be discussed in details. The
implementation of the Fuzzy logic will discussed. Then, the advantages
of the fuzzy logic over the classical classification techniques will
be discussed. 


Fuzzy Logic in Biological Control Systems 

 
 
The purpose of pattern recognition is a search for structure in data
which, we hope and assume, carries
information about the process generating them.
One of the means of achieving it is to use clustering where we want
to partition data into a
certain number of "natural" subsets where the elements of each set are
as similar as possible to each other and as different as possible from
those of the other sets.  Here, we shall discuss the fuzzy c-means
clustering algorithm and its variations including algorithms
which are able to recognize shape of clusters.  In addition, we
shall talk about cluster validity, i.e. how to check
whether the discovered structure of the data is the "correct" one.
 


Fuzzy Logic in Biological Pattern Recognicition

   

The purpose of the pattern recognition is to partition data into a
certain number of natural subsets where the elements of each set are
as similar as possible to each other and as different as possible from
those of the other sets. Here, we will discuss the fuzzy
clustering algorithms including fuzzy c-means, fuzzy c-shells
algorithm with classifier design.


Probabilistic Reasoning and of Uncertainty Managements with
Biomedical Applications


In medical applications, the relationships are often inexact and the 
conclusions are uncertain. Reasoning with such imprecise information,
the physician has to make a decision about the status of patient. For
these reasons, the use of probabilistic techniques to create the
medical expert systems which perform uncertainty reasoning is one
of the important research areas.

Although the need for handling of uncertainty in computer models has
been recognized since the beginning of computer modeling, early
attempts used mainly adhoc approaches with minimal theoretical
foundations. At the same time, theoretical foundations of fuzzy logic
and other techniques of approximate reasoning were under development.
For a number of years, these areas of research progresses separately
with little interaction. In the last decade theoretical developments
have been combined with practical implementations to provide systems
with sophisticated techniques for handling uncertainty. The evolution
of these methods will be discussed.

A number of approaches to the handling of uncertainty exist, including
certainty factors, probabilistic interference, Dempster-Shaffer
inference, fuzzy logic, and approximate reasoning. These methods will
be compared, contrasted and illustrated with concrete examples. A
number of fuzzy logic techniques will be illustrated, including fuzzy
numbers applied to test results, fuzzy matching to determine degree of
normalcy, the use of membership functions, methods for determination
of appropriate membership functions, and the types and uses of
linguistic quantifiers.



Fuzzy Expert Systems with Medical Applications


In the development of expert systems, a number of components must be
considered as sources of uncertainty. These include the knowledge
base, the patient case information, the reasoning process, and the
evaluation of results. Implications of uncertainty in each of these
areas will be discussed, along with alternative methods for dealing
with each.

Fuzzy logic techniques in expert systems will be illustrated in
detail, including the design of fuzzy knowledge bases, handling of
uncertainties in patient data. and implementation of fuzzy inference.
Participants will have an opportunity to use a fuzzy medical expert
system for analysis of chest pain to see how these techniques are
utilized in practice.




   Knowledge-based fuzzy expert systems will be compared with other
approaches, including connectionist systems, Bayesian systems,
and hybrid approaches.  Methods for defining membership functions
in knowledge-based systems will also be presented.


Fuzzy Logic in Nonlinear Systems with Biomedical Applications


Approaches for inclusion of chaotic data in expert systems will
be discussed, along with general issues in soft computing which
focus on approximate solutions to practical problems rather 
than on the development of precise mathematical models which
produce intractable solutions.


A number of the fuzzy logic approaches to nonlinear systems
will be illustrated for a number of medical applications including
diagnosis in cardiology, noninvasive detection coronary artery disease,

development of prognostic models in melanoma, and analysis of test
results in lung cancer.


SECOND DAY

 APPLICATIONS OF FUZZY LOGIC TO MEDICINE AND BIOLOGY

 Participation:      Dr. Metin Akay (Chair), Rutgers University
                     Dr. Donna Hudson, UCSF
                     Dr. Maurice Cohen, UCSF
                     Dr. Larry Hall, USF
                     Dr. Howard Jay Chizeck, Case Western
                     Dr. Patrick Eklund, Finland
                     Dr. Safwan Shah, Univ of Colorado, Boulder
                     Dr. Krzysztof Cios, Univ Of Toledo
                     Dr. Bhavin V. Mehta, Ohio University


FUZZY SETS ARE HELPFUL IN THE DIAGNOSIS OF CORONARY ARTERY STENOSIS
                      Krzysztof Cios


In the first part, we will discuss methods for defining fuzzy sets for the
recognition of coronary artery stenosis from post-exercise
planar thallium-201 scintigrams.  Probability density functions
are used to define fuzzy sets. Machine learning algorithms
are used to generate diagnostic rules.  These rules are then compared
for accuracy with rules specified by a nuclear cardiologist, using
sensitivities and specificities for each rule system.
We showed that we needed only 15 rules generated by the EXP machine
learning algorithm in order to  perform as well as 68
cardiologist-specified rules. 
 
The second part of the talk will describe a system in which fuzzy
generalized operators are aggregated with decision rules generated by the
machine learning algorithms EXP, CLILP2 and ALFS.  Results obtained
for the three sets of decision rules yield, on average, an accuracy of
above 90%.


    SEGMENTING BRAIN IMAGES WITH A FUZZY CLASSIFIER
               Lawrence O. Hall


The segmentation of magnetic resonance images of the brain into 
regions of interest is discussed.  A fuzzy hybrid connectionist, s
ymbolic instance based learning algorithm is used.   Fuzzy rules can
be generated after learning and the rules may  provide  information
about  important  inter- relationships among image features.
Continuous attributes are  fuzzified in this approach to learning.
The generated fuzzy rules may be used in Fuzzy-CLIPS expert system
tool for classification. Segmentation  results and issues are
discussed. The results show an excellent segmentation of a normal central
slice of the brain into gray matter, white matter and
cerebro spinal fluid.


   FUZZY ANALYSES OF BIOLOGICAL INFORMATION PROCESSING
                          Safwan Shan


	Neural information processing is mediated by complex
	inter-relationships amongst cells. Existing methods
	to carry out comprehensive evaluations of these
	interactions are constrained by properties intrinsic
	to neurobiological data: low mean cellular firing
	rates and aperiodic sequences of firing times. We
	have devised a technique to overcome these
	difficlties. Using fuzzy logic to represent cellular
	information processing characteristics,
        we examined the functional dynamics of
	multi-unit spiketrains. It was found that a fuzzy
	analog of a neural spike train enhance the
	opportunities to detect complex cellular
	interactions and together with ANN's provides a
	powerful environment withinn which to study neural
	information processing.


A NEURO-FUZZY MEDICAL DECISION SUPPORT DEVELOPMENT WORKBENCH
                        Patrik Eklund


   This work describes a laboratory and clinical support systems workbench, 
DiagaiD, based on an efficient transfer of patient data between health care 
professionals and clinical subsystems. The objectives of DiagaiD is to 
provide a knowledge elicitation tool, and to support automatic generation of 
stand-alone clinical decision support systems.
   Domain experts can utilize the workbench  in order to generate 
rule bases and graphical user interfaces, and integrating them into 
stand-alone systems. Thus the classical knowledge acquisition technique, 
heavily depending on close cooperation between domain experts and 
systems engineers, is replaced by fully automatic eliciation facilities.  
   DiagaiD has been installed in hospital use. The goal of this paper 
is to demonstrate the power of the DiagaiD workbench. 


  FUZZY LOGIC CONTROL FOR RESTORING MOVEMENT TO PARAPLEGIC INDIVIDUALS
                Howard Jay Chizeck



Limitations in the ability to stand, to negotiate curbing and stairs, and  
to walk reduce the physical access of paraplegic individuals to certain 
locations, restrict their employment opportunities and can lower their overall 
quality of life.  Existing functional neuromuscular stimulation (FNS) devices 
for locomotion in paraplegic, stroke and head injury patients have involved 
mostly open loop (pre-programmed) stimulation patterns, rather than feedback 
control.  This approach precludes correction for undesirable movements and 
generally requires excess stimulation of muscles.  Specific problems with 
existing FNS systems include limited independent locomotion (i.e., an 
assistant is needed when walking, for most subjects); variability of
motion; and a lack of compensation for fatigue and for changes in
slope and surface during FNS walking.  

A novel fuzzy-logic based structure has been developed that uses 
measurements obtained during one step to adjust the electrical stimulation 
during subsequent steps of the gait cycle.  The paraplegic subject receives 
stimulation (of up to 48 channels) from the stimulator and implanted
electrodes. 

Sensors measure physiological quantities during gait.  A fuzzy logic based gait
event detector determines which phase of gait the subject is currently
performing.  The rules for this gait event detector come from two sources. Some
rules are "hand crafted," are based common sense and experience of the
investigator.  Other rules  were derived using fuzzy 
system identification methods, on the basis of sensor measurements and visual 
observations.  This information is used, along with the sensor measurements, by
a gait evaluator which processes gait phase information (from the detector) and
sensor measurements (angles, forces) to determine the "quality" of the gait 
during each phase. A pattern adjuster then alters the stimulation, using fuzzy
logic. Sensor measurements used to detect gait events include: foot-floor 
contact force and center-of-pressure (insole-mounted transducers); ankle angles
(brace-mounted potentiometers); and knee angles (skin-mounted angle sensors).
Gait anomalies, such as foot drag, and changes to the external environment
(such as walking surface slopes) can be accurately detected using this
approach, and necessary adjustment to the locomotion stimulation
pattern can be made. By improving the quality, repeatability and
predictability of FNS gait, this work is a contribution to improving
the practicality of FNS locomotion systems. 


APPLICATION OF FUZZY ARTMAP FOR PREDICTION OF PROTEIN 3D STRUCTURE
                    Bhavin V. Mehta


Several techniques have been developed for protein secondary structure
prediction using different type of statistical and neural network techniques. 
In this paper, the Fuzzy ARTmap paradigm and the Probabilistic Network
algorithm is used.  The results of this new methods applied to
secondary structure prediction is compared with the existing
techniques like back error propagation method and the Chou-Fassman
statistical method.  
Fuzzy ARTmap which has been selected in predicting the secondary
structure of proteins because of its architectural and neurodynamical 
characteristics.is an incremental supervised learning algorithm which 
combines fuzzy logic and adaptive resonance neural network for the
recognition of pattern categories and multidimensional maps in
response to input vectors presented in arbitrary order.  
Its neurodynamics implements a new min-max learning rule which
conjointly minimizes predictive  error and maximizes code compression,
and therefore generalization. In addition, Fuzzy ARTmap is easy to
use. It has small number of parameters, requires no problem specific
crafting or choice of  initial weights, and does not get trapped in
local minima (specially with large  data sets that we are using).
Fuzzy ARTmap has been implemented in a simulator developed in-house
using the C language. 

The analysis of the results have demonstrated that the development 
of a multi-expert architecture with different representation schemes 
could be a better solution. This multi-expert architecture would
utilize rules based on statistical analysis and neural networks to grow with
examples. In addition, the utilization of additional information such as 
energy levels is being considered.


USE OF APPROXIMATE REASONING IN A HYBRID EXPERT SYSTEM 
              FOR CHEST PAIN ANALYSIS
                   Donna Hudson


A decision support system for the analysis of chest pain
in the emergency room environment will be discussed and
demonstrated.  The system consists of a rule-based 
component which uses techniques from approximate reasoning
to combine partial presence of symptoms and weighted
antecedents to determine rule substantiation.  This structure
permits fuzzy techniques to be incorporated into the decision 
making process at two stages:  in the development of the knowledge
base and in the direct use of the system by the end user.  
The weighted antecedents provide a more complex rule structure
in which the expert is able to incorporate information pertaining
to the relative importance of each contributing parameter.
The partial presence of symptoms allows nuances in the data to be
recorded by the user more accurately than with a simple yes/no response.
In addition to the rule-based component, a neural network model is
included which can be used to determine membership functions and
rule thresholds, or can be used directly as a decision making tool.
This is a functioning system which will be demonstrated on an IBM-
compatible portable computer.  Participants will have the 
opportunity to try the system individually.


   USE OF CHAOTIC ANALYSIS AS A MEANS OF QUANTIFYING LEVELS OF
      VARIABILITY IN THE ANALYSIS OF HOLTER R-R INTERVALS 
                          Maurice Cohen


Analysis of time series data is subject to problems
of nonlinear analysis, which can often lead to chaotic
states.  A method for analyzing 24-hour Holter data will
be discussed.  Analysis of these data sets in complicated
by the large number of points which are recorded in the
24 hour cycle, usually in excess of 100,000.  Although 
graphing procedures can provide pictorial evidence of the 
amount of variability in these data, the physical plotting
procedure can take several hours using an automatic plotter
and often results in unreadable plots if all points are plotted
on the same graph.  Through a continuous solution of the logistic
equation, a second-order difference plot has been developed which
demonstrates the degree to which these data points center around
the origin.  Again, because of the large number of points, a
numerical measure of central tendency will be presented, which
gives a value between 0 and 1, with 1 indicating no variability.
These measures can be used to develop membership functions which
can be used in conjunction with fuzzy models in either knowledge-based
systems or neural networks.




   NONINVASIVE DETECTION OF CORONARY ARTERY DISEASE USING
                 FUZZY MIN-MAX METHODS
                       Metin Akay

This study examines the use of fuzzy min-max method
for detecting coronary artery disease noninvasively by 
extracting useful information from the diastolic heart sounds
associated with coronary occlusions. It has been widely reported that
coronary stenoses produce sounds due to the turbulent blood flow in
these vessels. These complex and highly attenuated signals taken from 
recordings made in a soundproof room were detected and
analyzed to provide feature set. In addition, some physical
examination variables such as sex, age, body weight, smoking
condition, systolic and diastolic pressure were included in the feature vector.



More information about the Bionews mailing list