ArticlePDF Available

Newer applications of neuro-fuzzy systems for risk assessment and diagnostics in medicine

Authors:
  • Visoka tehnička škola strukovnih studija, Serbia, Subotica
  • Subotica Tech - College of Applied Sciences Subotica

Abstract and Figures

The application of neuro-fuzzy systems in the fields of medical sciences are increasing in parallel with the rapid evolution of diagnostic and treatment procedures in medicine. Contemporary literature provides us with many examples that this cooperation has brought many benefits in the fields of diagnostics and risk assessment in medicine. In Serbia, similar to those more developed countries, these systems have found their application from all branches from primary to tertial health care and in those institutions that work on more than one levels of healthcare. It seems that the well documented initial refuse towards the use of neuro-fuzzy systems by physicians starts lose its power.
Content may be subject to copyright.
Figure 1. Neuro-fuzzy information feedback system
Newer applications of neuro-fuzzy systems for
risk assessment and diagnostics in medicine
Dr. Andor Sagi*, Anita Sabo**, Tibor Szakáll**
**
Subotica Tech, Subotica, Serbia
peva@vts.su.ac.rs, saboanita@vts.su.ac.rs, szakall.tibor@gmail.com
Abstract The application of neuro-fuzzy systems in the
fields of medical sciences are increasing in parallel with the
rapid evolution of diagnostic and treatment procedures in
medicine. Contemporary literature provides us with many
examples that this cooperation has brought many benefits in
the fields of diagnostics and risk assessment in medicine. In
Serbia, similar to those more developed countries, these
systems have found their application from all branches from
primary to tertial health care and in those institutions that
work on more than one levels of healthcare. It seems that
the well documented initial refuse towards the use of neuro-
fuzzy systems by physicians starts lose its power.
I. THE ROLE AND PLACE OF NEURO-FUZZY
SYSTEMS IN MEDICAL SCIENCES
Medical diagnostics, despite the wide use of
international Latin nomenclature and disease classification
systems, is a field of evaluation of vague, human-like
categories. Risk assessment requires high degree
modeling, but on the same time must provide the
necessary environment of interaction with the human
factor.
To establish objective quality control the process must
be monitored by a multi-level feedback system operated
by a team of experts preferably of medical informatics,
social medicine, epidemiology and by a committee for
verifying the accuracy of the established diagnoses. These
factors are important not solely for the purpose of
satisfying the principles of Good Clinical Practice (GCP),
but also the terms defined by social medicine and the
country`s main development program.
These systems will need to be changeable as clinical
considerations may alter the existing categories. The
frequent lowering of tolerance levels for LDL cholesterol
provides us with excellent example. These changes are
often unforeseen in medical sciences so the speed of
changing these systems will come to foreground. In the
previous example the interaction of business interest of
antihyperlipemic drug manufacturers and changing
medical aims among others resulted in frequent
changes of these values.
II. MODERN
USES OF NEURO-FUZZY SYSTEMS
FOR RISK ASSESMENT AND DIAGNOSTICS IN
MEDICINE
Neuro-fuzzy systems, among other requirements, have
to meet not only the standards and expectations of
physicians, but their patients as well. An ample example
for this is the use of neuro-fuzzy systems by a group of
prominent scientists of the Annamalai Univerity. Their
work, published in 2005 is titled “An investigation of
neuro-fuzzy systems in psychosomatic disorders” in
which they state the possibility of interpreting the vague,
human-like symptoms of the hospitalized as well as
appropriately diagnose their psychic condition. I might
recollect that this is one of the most ambitious examples of
the efficacy of neuro-fuzzy systems, since it was able to
validly simulate the understanding of complex
pathophysiological laws to the extent of valid diagnoses
with its effectiveness approved physicians. It is for the
first time that a system that is based on artificial neural
networks actually used directly for the evaluation of
function of a real neural network.
Mental diseases and psychosomatic disorders are
closely determined with the structural architecture and the
conductive abilities of this biological network. In the age
of examining the human brain the simulation of such a
system becomes imperative as medical ethics greatly
hinders the possibilities of legal experiments on humans.
The electric charge in the neurons is generated by a
sodium influx and potassium efflux which depolarizes the
cell. Complex, additional biochemical reactions make
possible the flow of this current. The speed of current
propagation as well as its strength and other properties are
now clearly defined, opening the path for making
appropriate models. Current conduction by chemical
synapses and the properties of the certain
neurotransmitters is also a subject of knowledge. The
number of possible connections if billion times billion, but
the histological and neurological properties of the human
brain still provide ample ground for modeling purposes.
Many mental disorders are mostly due to abnormal
production or lack of certain neurotransmitters or their
quantitive proportions. In these examples no higher
structural analysis is required. On the other hand in
epilepsy the abnormal excitation of pathologic neuron
groups provide the clinical picture. Since these
phenomena can be measured by EEG the neuro-fuzzy
system can be filled up with data intervals necessary to
classify vague signals. Epilepsy can be triggered with
many external influences which patients can describe only
subjectively and therefore not precisely. These effects
- 617 -
SISY 2010 • 2010 IEEE 8th International Symposium on Intelligent Systems and Informatics • September 10-11, 2010, Subotica, Serbia
978-1-4244-7395-3/10/$26.00 ©2010 IEEE
Figure 2. Neuro-fuzzy modeling of internal organs for surgical
simulators
could be intense or rapidly changing visual or auditive
impulses.
The role of simulation becomes evident in the so-called
surgical simulators too. Surgeons use these simulators to
practice new surgical methods before actually applying
them on patients. Apart from realistic 3d modeling they
also need to recreate the feel and consequences of
manipulating with elastic tissue. The tissue`s level of
elasticity was often left without simulation creating an
unreal practicing environment. In the modern surgical
simulators the modeling and realization is achieved using
neuro-fuzzy systems. The Institute for Medical
Informatics, Technical University of Braunschweig, and
Institute for Knowledge and Language Processing,
University of Magdeburg together developed the plan for
such a sytem [2,1998].
It is also desirable that the modeling respects at least the
main histological layers of these organs, required for
example at microsurgical or some gynecologic or
otorhinolaryngological operations. During these
procedures the physician is using a sort of microscope and
usually interferes with the integrity of very thin
histological layers, leaving more healthy tissue intact. The
evolution of these multi-layer systems is at the stage of
planning, but with the speed of evolution of this branch of
medical informatics it is considered soon to be realized.
There is on the other hand some concern that these
simulations require more expensive computers. The most
important and the most common aspects of possible
application of neuro-fuzzy systems are in primary health
care.
Besides the anamnesis and physical examination,
laboratory diagnostics is usually among the most often
diagnostic method in the inventory of a general
practitioner, but is essential part of any preventive branch
of medicine. The anamnesis and physical examination is
written in the patient’s records and their interpretation and
control is relatively objective. The laboratory results on
the contrary only contain the measured value and the
intervals in which these values are to be regarded as
physiological. The patient is deprived from any insight of
his condition, while doctors can also interpret them
sometimes too deliberately (since often there are many
proposed interpretations in medical literature).
Nevertheless this system cannot be also effectively
integrated in a bigger, modern neuro-fuzzy based one for
diagnostics or risk assessment.
A possible solution was provided for numerous values
by the J.W. Goethe University of Frankfurt [5,2000].
Among others fuzzy categories where defined for the
following: mean corpuscular erythrocytical volume,
alkalic phosphatase, glutamat-pyruvat-transaminase in
serum, glutamat-oxalacetat-transaminase in serum
gammaglutamyltranspeptidase, and for the number of
daily consumed alcohol portions. The significance of this
research is pointed out in the Proceedings of European
Symposium on Artificial Neural Networks, ESANN 2000,
Elsevier Publ., 2000, pp. 201-206 with these words: “The
decision and diagnostic power of neural networks are
needed in many tasks of everyday life. Successful
example applications show that especially in medicine
human diagnosis abilities are significantly worse than
those of neural diagnosis systems.“
The electrocardiogram is also a common device in
diagnostics and cardiovascular risk evaluation in all levels
of healthcare. The proper interpretation of the
electrocardiogram is time consuming for every physician
since there are 12 leads to consider, not to mention their
comparison, becoming impractical for many branches of
healthcare like intensive care units, coronary care units,
ambulance or any physician providing life support. Many
successful attempts have been made for that matter to
construct an ECG device which also establishes a
diagnose. These ECG devices are using fuzzy logic and
are relatively common in Serbia. The disadvantage of
these systems is their reliability which varies from
manufacturer to manufacturer, and therefore cannot be
trusted by physicians. The great responsibility of
administrating the proper drug in certain acute situations
and the general flaws of these devices prompt doctors
toward manual diagnosis making regardless of the
situation. This is one of the compelling reasons why
diagnosing ECG devices are not significantly preferred to
the ones who do not use fuzzy logic.
Many groups of authors recently published works
independently on the possibilities of applying neuro-fuzzy
systems to enhance the capabilities of artificial ECG
interpretation. It is encouraging that these studies are
conducted even in this region like in the Dept. of Applied
Electronics and Information Engineering, Polytechnic
University, Bucharest, Romania [6,2003]; providing
possibilities for Serbian scientists possibilities of
collaboration in this field. Romania published its results
under the title: “A Neuro-Fuzzy Approach to
Classification of ECG Signals for Ischemic Heart Disease
Diagnosis”. They state that the rate of correctness for
ECG interpretation for Ischemic Heart Diseases is already
100% in the examined number of patients.
In 2007 a group of scientists from various Chinese
universities discussed the possibility of describing
complex hormonal regulation mechanisms using the
neuro-fuzzy technology. Endocrinology is a particularly
complex branch of medicine, since the adjustment of the
desired hormonal status is the function of many organ
systems, the brain, the numerous endocrine glands, blood
hormone levels, the concentration of sodium in the blood
or osteopoesis are only a few from the various interacting
elements.
A typical neuroendocrine mechanism can be described
with the following generalized algorithm, the knowledge
of which is the basis of making efficient neuro-fuzzy
systems in endocrinology. The hypothalamus produces
releasing hormones which are transported to the glandula
pinealis. The glandula pinealis releases a so called
endocrine gland stimulating hormone. This stimulating
hormone influences gland cell behavior and biochemical
processes, allowing the gland to produce the desired
product. The so produced hormones are directly injected
in the blood flow and travel to the target cells.
A. Sagi et al. • Newer Applications of Neuro-Fuzzy Systems for Risk Assessment and Diagnostics in Medicine
- 618 -
Figure 3. Schematic block diagram of the computational hand
grip assessment tool
In certain diseases the hormone could get blocked while
residing in the blood flow. The presence of certain
receptors on the target cells are usually microscopically
connected with certain proteins and/or induce biochemical
chain reactions altering the host’s metabolism. The
presence, absence or quality of these receptors is the main
factor in many widely common diseases such as Diabetes
type II, from which approximately 8% of the population
suffers. There are many feedback mechanisms that inform
the various members of hormonal homeostasis of the
hormonal status, lowering or otherwise adjusting their
function. All of this is underlined with a firm histological
basis, but the scope of this discussion doesn`t permit
further discussion on it, nevertheless the presented system
is also sufficiently applicable for modeling. Frequently
tumors of the neuroendocrine organs alter this fine-tuned
regulation profoundly, which due to the high prevalence
of these neoplasms - is also should be part of a more
sensitive neuro-fuzzy based simulation.
Neuro-fuzzy systems are more and more accepted in
interpreting the results of ophtalmolmological
examination. Since ophthalmology a largely visual branch
of medical science, various interpretations and judgments
are possible for certain conditions. The universities of
Madrid and Tenerife have worked together to make a
neuro-fuzzy system for automatic interpretation of the
visual field [3, 2001]. The scientists stated that it achieved
high accuracy and was generally able to substitute the
physician in the examined cases. Similarly to these
examples, almost every medical field in principle that
requires any sort of interpretations of vague, human-like
categories is open for introducing neuro-fuzzy
interpretators.
There is a vast field of possibilities in introducing these
systems in the process of physical therapy and
rehabilitation of patients. As an example of such a use of
neuro-fuzzy system a particular experiment will be
demonstrated. Scientists from the University of Malaya
tried to objectively express the strength of human grip as a
means of a more precise evaluation of the progress of
rehabilitation as well as neural damage [1, 2006]. Test
results clearly showed that an Adaptive Network-based
Fuzzy Interference System is “a robust and adaptive tool
to recognize human hand grip strength patterns. “. This
system has also an immensely accurate tool for diagnostic
purposes, mainly diseases of the musculoskeletal system,
for example almost all the forms of myopathias. The
professors are constantly improving the recognition
accuracy; it is currently about 85%. This solution is also a
valid tool for post-operative monitoring and healthcare. It
is also important that the signals are “calibrated, filtered
and stored in a database for retrieval and querying
purposes.”, and that these data are presented in an
acceptable way to the physician.
The strength of the hand grip is detected using strain
gauges, and then an electrical is produced, registered by
the Labview software. Then it is transferred to the data
classification unit. Personal computers enable practical
storage of the data, thus creating a comprehensive patient
record and enables comparison so monitoring the
effectiveness of treatment in the function of time. A brief
schematic block diagram of this computational hand grip
assessment tool is given below. I leave out the particular
technical details, because there are other prominent
projects for the evaluation of human hand grip strength,
which achieve similar results by using different
specifications.
III. NEURO-FUZZY
SYSTEMS IN MEDICINE
CHALLENGES AND FRONTIERS
The hand grip assessment tool`s significance is clear
from the fact that many authors on the application of
neuro-fuzzy systems agree that this point of
communication is one of the most problematic and
controversial in these systems. This problem lies mostly in
the complexness of medical, international Latin
terminology and language which can be hardly generated
by artificial intelligence. The debates were mostly
centered about what the nature and detail of the
presentation of the collected data should be. Artificial
intelligence often presented these data by a not precise
computer generated language or in a way that was
understandable only for the technicians, therefore not
practical. This was one of the main reasons of refusal by
the professional healthcare staff.
One of the most comprehensive medical textbooks on
the subject of neuro-fuzzy systems nowadays is “Fuzzy
and Neuro Fuzzy Systems in Medicine” edited by Horia-
Nicolai Teodorescu et al. In this textbook, apart of
defining the current challenges and status of this
technology, the author also presents us case-studies based
on medical experiments [4,1991]. In the following lines I
mention some of the biggest and most ambitious projects
besides the ones I already elaborated in this work. Though
their mention is more due to a simple list, it still validly
represents in which branches of medical science was the
adaptation of neuro-fuzzy systems was most wide.
There was, as I mentioned before, many attempts to
understand the brain as a fuzzy system and modeling its
functions; identifying its state or forecasting acute
pathology. Brain surgery was performed and
segmentation of neural tumors this way. Many
experiments were conducted on diagnosis and treatment
for diseases of the myocardium (mostly for ischemic heart
disease) or blood pool. The establishment of an expert
system in intensive care diagnostics (for addressing
conditions such as heart failure or shock) has reported to
be effective in lowering patient mortality. Many practical
methods are proposed for this purpose, but according to
Teodorescu they are of ”comparable complexity and have
similar performances”.
Analyzing EEG signals in neurology is now possible
using various technical solutions: time frequency analysis,
multiscale decomposition by the fast wavelet transform,
- 619 -
SISY 2010 • 2010 IEEE 8th International Symposium on Intelligent Systems and Informatics • September 10-11, 2010, Subotica, Serbia
multichannel model based and decomposition by
matching pursuit. Their classification is usually conducted
by one of the following methods: by an unsupervised
optimal fuzzy clustering algorithm or the weighted fuzzy
k-means algorithm.
Combination of neuro-fuzzy systems with wavelet
analysis brought many possibilities to exist among various
branches of medical imaging technology. SPECT of the
ventricular myocardium is one of the most stunning
examples of successful practical implementation of this
combination. MR image analysis is also a prominent
example for the use of neuro-fuzzy systems in medical
imaging, tested for unsupervised brain tumor
segmentation and labeling. The edge of the tumor mass
could be determined more precisely then with other pre-
existent method, allowing more precise data for surgical
purposes. This system is tuned by the most possible
accuracy (since it is for surgical purposes) on two levels,
similar to other neuro-fuzzy systems: on the level of
physicians and on the fuzzy categories which is also called
“fine tuning”.
So far we have discussed strictly the medical
applications of neuro-fuzzy systems, although it is
interesting that stomatology also began to benefit from
them. Stomatology is closely interwoven with a branch of
medical science, maxillofacial surgery and therefore its
worth to mention that occlusion analysis and analysis of
masticatory function is already realized using this
technology. Besides maxillofacial surgery, a pediatric
advantage is defined, namely the reducing of the
necessary amount of X-ray examination of children, that
way lowering the accumulation of radiation.
Hemodynamic control during anesthesia by a neuro-
fuzzy system has proved to be a problem which solution is
among the future challenges of medical informatics. High
level of uncertainty, the complexity of control and
constantly changing physiological parameters did not
allow successful results.
While many attempts have been made to design a
system that could work without medical supervision, these
systems prove themselves not stable and still have to be
controlled by physicians.
IV. C
ONCLUSIONS
Despite promising results, the use of neuro-fuzzy
systems in medicine still faces certain general problems
regardless of the field of medical science. It is already
mentioned that many of these systems (surgical simulators
for example) require more sophisticated and therefore
more expensive hardware, therefore must be subject to
evaluation by social medicine. The use of these systems
provide many, but different advantages on all three levels
of healthcare, as well as for those institutions operating on
multiple levels, therefore it must be controlled by the
countries healthcare development program mainly from
economical perspective.
Hardware minimization or optimization between
architecture and design of these devices are also listed
among the most discussed problems. Most authors agree
that the main constrains of investigation of the
effectiveness of a neuro-fuzzy system is evaluation of the
systems autonomy, reliability and the precision of
computation.
R
EFERENCES
[1] A. Hafiz, S. Anandan, T. Kamarul “Recognition of Human Grip
Strength Patterns Using Neuro Fuzzy Technique Proceedings of
the Third International Conference on Artificial Intelligence in
Engineering & Technology November 22-24, 2006, Kota
Kinabalu, Sabah, Malaysia
[2] A. Radetzkya, A. Nürnbergerb, D. P. Pretschnera “The simulation
of elastic tissues in virtual medicine using neuro-fuzzy systems”,
SPIE, Bellingham WA, ETATS-UNIS, 1998
[3] E. Carmona, J. Mira, J. G. Feijoo, M. G. de la RosaNeuro-Fuzzy
Nets in Medical Diagnosis: The DIAGEN Case Study of
Glaucoma” Proceedings of the 6th International Work-Conference
on Artificial and Natural Neural Networks: Bio-inspired
Applications of Connectionism-Part II ,pp. 401 409, 2001
[4] H N Teodorescu, Horia Nicolai L Teodorescu, Abraham Kandel
“Fuzzy and Neuro-Fuzzy Systems in Medicine”, CRC Press, 1991.
[5] R. Brause, F. Friedrich “A Neuro-Fuzzy Approach as Medical
Diagnostic Interface” Proc. European Symposium on Artificial
Neural Networks ESANN 2000, Elsevier Publ., 2000, pp. 201-206
[6] VE Neagoe, IF Iatan, S. Grunwald “A neuro-fuzzy approach to
classification of ECG signals for ischemic heart disease
diagnosis.” AMIA Annu Symp Proc. 2003:494-8.
A. Sagi et al. • Newer Applications of Neuro-Fuzzy Systems for Risk Assessment and Diagnostics in Medicine
- 620 -
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
This work presents an approach to the automatic interpretation of the visual field to enable ophthalmology patients to be classified as glaucomatous and normal. The approach is based on a neuro-fuzzy system (NEFCLASS) that enables a set of rules to be learnt, with no a priori knowledge, and the fuzzy sets that form the rule antecedents to be tuned, on the basis of a set of training data. Another alternative is to insert knowledge (fuzzy rules) and let the system tune its antecedents, as in the previous case. Three trials are shown which demonstrate the useful application of this approach in this medical discipline, enabling a set of rule bases to be obtained which produce high sensitivity and specificity values in the classification process.
Article
Full-text available
The paper focuses on the neuro-fuzzy classifier called Fuzzy-Gaussian Neural Network (FGNN) to recognize the ECG signals for Ischemic Heart Disease (IHD) diagnosis. The proposed ECG processing cascade has two main stages: (a) Feature extraction from the QRST zone of ECG signals using either the Principal Component Analysis (PCA) or the Discrete Cosine Transform (DCT); (b) Pattern classification for IHD diagnosis using the FGNN. We have performed the software implementation and have experimented the proposed neuro-fuzzy model for IHD diagnosis. We have used an ECG database of 40 subjects, where 20 subjects are IHD patients and the other 20 are normal ones. The best performance has been of 100% IHD recognition score. The result is exciting as much as we have used only one lead (V5) of ECG records as input data, while the current diagnosis approaches require the set of 12 lead ECG signals!
Conference Paper
In contrast to the symbolic approach, neural networks seldom are de- signed to explain what they have learned. This is a major obstacle for its use in everyday life. With the appearance of neuro-fuzzy systems which use vague, human-like categories the situation has changed. Based on the well-known mechanisms of learning for RBF networks, a special neuro-fuzzy interface is proposed in this paper. It is especially useful in medical applications, using the notation and habits of physi- cians and other medically trained people. As an example, a liver disease diagnosis system is presented.
Article
this paper. It is especially useful in medical applications, using the notation and habits of physicians and other medically trained people. As an example, a liver disease diagnosis system is presented
Article
To improve the benefit of surgical simulators for education and research a visual convincing modeling of the operation scenario and the involved tissues is not sufficient. It is also necessary to simulate the deformation and resulting inner forces of tissue under influence of external forces caused by, for example, medical instruments or gravity. In this paper, we present a hybrid neuro-fuzzy system, which was designed for the description and simulation of tissues. The neuro-fuzzy system can be used to simulate the physical behavior like stiffness, viscosity and inertia of deformable or elastic tissues in surgical simulation. The parameters of a physical model or prior expert knowledge in the form of linguistic terms can be used to initialize the network parameters. Using a neural network structure, local changes to the system like cuts or ruptures can be performed during simulation. As an application example, some simulation results in the area of gynaecological laparoscopy are given....
Recognition of Human Grip Strength Patterns Using Neuro Fuzzy Technique
  • A Hafiz
  • S Anandan
  • T Kamarul
A. Hafiz, S. Anandan, T. Kamarul "Recognition of Human Grip Strength Patterns Using Neuro Fuzzy Technique " Proceedings of the Third International Conference on Artificial Intelligence in Engineering & Technology November 22-24, 2006, Kota Kinabalu, Sabah, Malaysia
Fuzzy and Neuro-Fuzzy Systems in Medicine
  • H N Teodorescu
  • Horia Nicolai
  • L Teodorescu
  • Abraham Kandel
H N Teodorescu, Horia Nicolai L Teodorescu, Abraham Kandel "Fuzzy and Neuro-Fuzzy Systems in Medicine", CRC Press, 1991.
The simulation of elastic tissues in virtual medicine using neuro-fuzzy systems
  • A Radetzkya
  • A Nürnbergerb
  • D P Pretschnera
A. Radetzkya, A. Nürnbergerb, D. P. Pretschnera "The simulation of elastic tissues in virtual medicine using neuro-fuzzy systems", SPIE, Bellingham WA, ETATS-UNIS, 1998