VISVESVARAYA TECHNOLOGICAL UNIVERSITY

JnanaSangama, Belgaum-590 014

2013 2014
Report on A Review on Biomedical
Applications of Artificial Neuron Networks.

MASTER OF TECHNOLOGY

By:
Bindushree Yadav.M
nd
2 semester, M. Tech, Biomedical Signal Processing and Instrumentation
R. V. College of Engineering, Bangalore

Department of Instrumentation Technology

R. V. College of Engineering,
Bangalore 560059

A Review on Biomedical Applications of Artificial

Neuron Networks.
Bindushree Yadav M Pavithra M N Priyanka K B
2nd semester, MTech, BMSP&I,
Department of Instrumentation Technology,
R.V College of Engineering,
Harsha, Assistant Professor,
Department of Instrumentation Technology,
R.V College of Engineering,
Bangalore,Karnataka,India
bindu.spandu@gmail.com
pavithjramn@gmail.com
priyankakbit42@gmail.com
harsha@rvce.edu.in
computation, learning ability, generalization ability,
adaptively, inherent contextual information processing, fault
tolerance, and low energy consumption. It is hope that devices
based on biological neural networks will possess some of
these desirable characteristics. Modern digital computers
outperform humans in the domain of numeric computation
and related symbol manipulation. However, humans can
effortlessly solve complex perceptual problems at such a high
speed and extent as to dwarf the worlds fastest computer.
Why is there such a remarkable difference in their
performance? The biological neural system architecture is
completely different from the von Neumann architecture. This
difference significantly affects the type of functions each
computational model can best perform.

Abstract--This paper presents a review of applications of artificial

neural networks in biomedical engineering area. Artificial neural
networks in general are explained; some limitations and some
proven benefits of neural networks are discussed. Use of artificial
neural network techniques in various biomedical engineering
applications is summarized.
Case studies are used to
demonstrate the efficacy of artificial neural networks in this
area. The paper concludes with a discussion of future usage of
artificial neural networks in the area of biomedical engineering.
Keywords: Biomedical applications, Breast cancer, ANN, Sleep
analysis, Hemodialysis, Skin diagnostics

I.INTRODUCTION

Numerous efforts to develop intelligent programs based on

von Neumanns centralized architecture have not resulted in
general-purpose intelligent programs. Inspired by biological
neural networks, ANNs are massively parallel computing
systems consisting of an extremely large number of simple
processors with many interconnections. ANNs models attempt to use some organizational principles believed to be
used in the human.

An artificial neural network is a powerful, versatile tool.

Artificial neural networks have been successfully used in
various applications such as biological, medical, industrial,
control
engendering,
and
software
engineering,
environmental, economical, and social applications. The high
versatility of artificial neural networks comes from its high
capability and learning function. It has been theoretically
proved that an artificial neural network can approximate any
continuous mapping by arbitrary precision. Desired
continuous mapping or a desired task is acquired in an
artificial neural network by learning.

II.CASE STUDIES
A. A Case Study: Breast Cancer Problem Domain

Numerous advances have been made in developing intelligent

systems, some inspired by biological neural networks. Researchers from many scientific disciplines are designing
artificial neural networks to solve a variety of problems in
pattern recognition, prediction, optimization, associative
memory, and control. The long course of evolution has given
the brain many desirable characteristics not present in von
Neumann or modern parallel computers. These include:
massive parallelism, distributed representation and

To demonstrate the effectiveness of neural networks in

biomedical engineering domain, experiments were carried
using ANN on breast cancer database. This database contains
instances of various breast cancers in several patients. The
target of the Breast Cancer database is to distinguish between
the benign and malignant type of cancer according to nine
attributes such as Clump thickness, Cell size, Cell shape,
Adhesion, Bare nuclei, Nucleoli, etc. The data set has 699
instances from which 16 instances were removed due to

missing information about the Single epithelial cell size

attribute. All the eliminated 16 patterns were instances of
benign type that already has a major distribution in instance
space. The resulting database contains 683 attributes, 444 of
them are of benign type and the remaining 239 are of
malignant type of cancer. To reduce learning complexity in
neural networks and assist in understanding the dependencies
among attributes and target concepts, we discretise
(categorise) attributes and use the sparse-coding
representation. Each value of a discrete attribute with n
possible values is represented by an n-bit binary string, with
only one bit carrying a value of one corresponding to the
attribute's value. For example, cell size is a feature that has
three values small, medium, large. This will be converted into
three binary features as size_small, size_medium, and
size_large representing the sparse-coding of {1 0 0}, {0 1 0},
and {0 0 1} respectively. This type of coding resulted in the
input layer with 90 nodes. To avoid the initial guess of neural
network architecture, we use the cascade correlation algorithm
one of the methods for incrementally building a feedforward
network that starts with an input and an output layer with no
hidden units. This algorithm constructs the network by
initially training a two layer (input and output nodes) model
only, and then gradually adding hidden nodes until an
acceptable overall network is achieved. The goal is to develop
a small size feedforward ANN with sigmoidal nodes that
properly classifies the training and unseen examples. The
Breast Cancer problem domain utilizes a 5-fold CV scheme to
produce the ANN solutions. The best network obtained has a
size of 90 input nodes, 2 hidden nodes and 1 output nodes.
Training of the neural network on this database means that the
resulting model should be able to diagnose an individual. The
resulting model must mimic the relationship among
physiological variables (such as Clump thickness, Cell size,
Cell shape, Adhesion, Bare nuclei, Nucleoli, etc, that we have
used for training) even at different physical activity levels. If a
model is adapted to an individual (pattern used for testing the
network), then it should be able to correctly predicting the
medical condition of that individual. When training of the
network has ceased, the root mean square error (RMSE) is
reduced to 0.0054. The trained network was able to
completely recognise the benign and malignant types of
cancer. From the patterns that were used for training, a 100%
accurate classification was achieved. While generalising with
the unseen patterns, the RMSE was 0.1887 (when tested on
239 unseen individuals, only 7 of them were incorrectly
predicted), thus yielding a high accuracy of the trained
network. This small experiment shows that the neural
networks trained on the breast-cancer problem database were
capable of predicting the new unseen cases with a high
accuracy. This demonstrates that neural networks can be
successfully applicable to biomedical engineering domain.

defined: movements, awake, REM sleep and sleep stages 1, 2,

3 and 4. The different signals (2 EEG, 1 EOG and 1 EMG
channels) were segmented into 2 s intervals. Each 30 s epoch
was characterized by 17 parameters (which are the 17
components of the input vector) obtained by averaging 15
consecutive estimates. These estimates were: EEG relative
power in the different bands delta (04 Hz), theta (48 Hz),
Alpha (813 Hz), Beta1 (1322 Hz), Beta2 (2235 Hz); total
power of the EEG (035 Hz); ratio of powers delta: theta;
ratio of powers alpha:theta; mean frequency and dispersion of
EEG spectral density; EOG relative power in the band (04
Hz); total power of EOG spectral density; mean frequency and
dispersion of EOG spectral density; total power of EMG
spectral density; mean frequency and dispersion of EMG
spectral density.
The neural network model used was a multilayer perceptron
trained with the classical backpropagation algorithm
(Rumelhart et al., 1986). The input layer (17 cells) was fully
connected to the hidden layer (10 cells) which was fully
connected to the output layer (6 cells). Using 11 all-night
recordings, agreement between neural network results and the
consensus of ten human experts reached 80.6%. In a second
study (Schaltenbrand et al., 1996), the performance of the
multilayer perceptron was tested on 20 healthy subject records
and on 40 patient records (20 depressive patients and 20
insomniac patients treated with a benzodiazepine). Agreement
between one visual scorer and the connectionist classification
was 84.5% for healthy subject records, 81.5% for depressive
and 81.0% for insomniac patient records.
C. A case study :Artificial neural network structure for
skin disease diagnostic
The design and architecture of ANN selected for the skin
diseases is based on the feed - forward network. This means
that the artificial neurons are organized in layers, and send
their signals forward, (i.e., from input to output) and then the
errors are propagated backwards (see, fig. 1). The network
receives input symptoms by neurons in the input layer, and the
output of the network is given by the neurons on an output
layer. There may be one or more

B. A case study: Schaltenbrands system (human sleep

analysis)
The system developed by Schaltenbrand et al. (1993) was
dedicated to human sleep analysis. Seven classes were

Figure 1: ANN for skin disease diagnostic

intermediate hidden layers. The back propagation algorithm

uses supervised learning, which means that we provide the
algorithm with examples of the inputs and outputs we want
the network to compute, and then the error (difference
between actual and expected results) is calculated. The idea is
to reduce this error, until the ANN learns the training data.
The training begins with random weights, and the goal is to
adjust them so that the error will be minimal. The activation
function of the artificial neurons in ANNs implementing the
backpropagation algorithm is given as follows (Haykin,
1999):

(1)
The output function uses the sigmoidal function:

Table 1. Truth table for the skin disease firing rule

formulation trick, which also requires some higher level
knowledge of the disease for which rule is formulated. The
rule provides a greater degree of flexibility for determining
whether scabies is a skin disease or not. This is particularly
true for a case where the patient has used cream to make the
skin less scaly. Omission of the scaly symptom will not affect
the decision of the neural network in determining whether to
suspect scabies or not. This is a more realistic situation which
definitely makes the neural network more desirable. The firing
rule gives the neuron a sense of similarity and enables it to
respond sensibly to symptoms not seen during training.

(2)
We defined the error function for the output of each neuron as:

(3)
The weights are adjusted using the method of gradient
descent:

D. A case study: Usefulness of artificial neural networks

to predict follow-up dietary protein intake in
hemodialysis patients

(4)

1) Background: Artificial neural networks (ANN) represent a

promising alternative to classical statistical and mathematic
methods to solve multidimensional nonlinear problems. The
aim of the study was to verify, by comparing the performance
of ANN with that of experienced nephrologists, whether ANN
are useful tools in hemodialysis to predict the follow-up (=1
month after the observation used for the prediction) dietary
protein intake (PCR), and whether their performance is
influenced by the size of the population and by the data pool
used to built the model.

Where: xi are the inputs, wij are the weights, Oj ( x , w ) are

the actual outputs, dj are the expected outputs and - learning
rate. For the skin diagnostic system, we defined the following
firing rules: take a collection of symptoms for a disease to a
node, the presence of which causes it to fire (the 1-taught set
of patterns) and the absence prevents it from firing (the 0taught set). If there is a tie in symptoms, then the ANN
remains in the undefined state (1/0). For instance, in scabies,
which have 4 - input symptoms, the neuron is taught to fire
(i.e., output 1), when the input (tiny bumps-X1, itching-X2,
scaly-X3 and on fingers-X4) is 1111 or 0111 or 1101 and will
not fire ((i.e., output 0) when the input is 0000 or 1010 or
1110. The firing rule is depicted in Table 1. It is important to
note that the decision concerning each firing rule was derived
directly from the Scabies condition. This condition specifies
that: Scabies occurs only on the hand has to be Itchy. These
conditions are weighed higher than other conditions. The
values 1111 or 0111 or 1101, implies that all conditions must
be met or at least there is either tiny bumps or scaly skin to
meet the 50% threshold mark required for firing. This is a
very important rule

2) Method: A combined retrospective and prospective

observational study was performed in two Swiss dialysis units
(84 chronic hemodialysis patients, 500 monthly clinical
observations and biochemical test results). Using
mathematical models based on linear regressions to evaluate
the variables, ANN were built and then prospectively and
interinstitutionally compared with the ability of six
experienced nephrologists to predict the follow-up PCR.
3) Results: ANN compared with nephrologists gave a more
accurate correlation between estimated and calculated followup PCR (P<0.001). The same superiority o fANN was also

seen in the ability to detect a follow-up PCR <1.00 g/kg/day

expressed as a percentage of correct predictions, sensitivity,
specificity, and predictivity. The interinstitutional performance
of the ANN is positively influenced by the size and the
variability of the population used to build the mathematical
model.

[1]

R. Nayak, L. C. Jain and B. K. H. Ting, Artificial Neural

Networks In Biomedical Engineering: A Review, Proceedings
Asia-Pacific Conference on Advance Computation,2001.

[2]

Khadijeh karamzadeh, Hadis Askarifar, Hamid

Moharrami,Biomedical Engineering Based Artificial

4) Conclusion: The use of ANN significantly improves the

ability of the experienced nephrologist to estimate and to
detect an unsatisfactory (<1.00 g/kg/day) follow-up PCR. The
size of the population selected to build the ANN is critical for
his performance.

[3]

Neural Network, International Journal of Scientific &

Engineering Research, Volume 4, Issue 6, June-2013.

[4]
[5]

III.CONCLUSION
[6]

It is obvious from our study that neural networks have been

used successfully in many areas in biomedical engineering. It
is obvious from literature that researchers have used neural
networks as computational tools, modeling tools as well as
human brain mimicking tool. Some potential biomedical
engineering fields where neural networks can be applied in
future are electrophysiology, biomaterials, biotechnology,
biosensors, modelling, instrumentation, rehab engineering,
medical analysis, prothetic, informatics, imaging, clinician,
biomechanics, computers devices.

[7]

REFERENCES

Claude Robert , Christian Guilpin , Ayme Limoge

Review of neural network applications in sleep research,
Journal of Neuroscience Methods 1998.
Bakpo, F. S.and Kabari, L. G, Diagnosing Skin Diseases Using
an Artificial Neural Network, IEEE.
Kenji Suzuki ,Artificial Neural Networks Methodological
Advances And Biomedical Applications, Copyright 2011
InTech
Nikolai Petrovsky, Soh Khum Tam, Vladimir Brusic, Graeme
Russ, Luis Socha and Vladimir B. Bajic Use of Artificial
Neural Networks in Improving Renal Transplantation
Outcomes, Graft 2002.
Luca Gabutti, Michel Burnier, Giorgio Mombelli, Francesca
MalE, Lisa Pellegrini, And Claudio Marone Usefulness of
artificial neural networks to predict follow-up dietary protein
intake in hemodialysis patients, Kidney International, Vol. 66
(2004), pp. 399407