International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue (1)

Editor in Chief Professor João Manuel R. S. Tavares

International Journal of Biometrics and
Bioinformatics (IJBB)

Book: 2009 Volume 3, Issue 1
Publishing Date: 30-02-2009
Proceedings
ISSN (Online): 1985-2347

This work is subjected to copyright. All rights are reserved whether the whole or
part of the material is concerned, specifically the rights of translation, reprinting,
re-use of illusions, recitation, broadcasting, reproduction on microfilms or in any
other way, and storage in data banks. Duplication of this publication of parts
thereof is permitted only under the provision of the copyright law 1965, in its
current version, and permission of use must always be obtained from CSC
Publishers. Violations are liable to prosecution under the copyright law.

IJBB Journal is a part of CSC Publishers
http://www.cscjournals.org

©IJBB Journal
Published in Malaysia

Typesetting: Camera-ready by author, data conversation by CSC Publishing
Services – CSC Journals, Malaysia

CSC Publishers

Table of Contents

Volume 3, Issue , Febuary 2009.

Pages

1-9 Reducing Process-Time for Fingerprint Identification System
Chander Kant, Rajender Nath.

10 - 18 Designing an Artificial Neural Network Model for the Prediction of
Thrombo-embolic Stroke
Shanthi Dhanushkodi, G.Sahoo , Saravanan Nallaperumal

International Journal of Biometrics and Bioinformatics, (IJBB), Volume (3) : Issue (1)

Chander Kant & Rajender Nath

Reducing Process-Time for Fingerprint Identification System

Chander Kant ckverma@rediffmail.com
Lecturer, Department of Computer Science and
applications K.U.
Kurukshetra, Haryana (INDIA)

Rajender Nath rnath_2k3@rediffmail.com
Reader, Department of Computer Science and
applications K.U.
Kurukshetra, Haryana (INDIA)

ABSTRACT

Fingerprints are the most widely used biometric feature for person identification and
verification in the field of biometric identification. Fingerprints possess two main types of
features that are used for automatic fingerprint identification and verification: (i) Ridge
and furrow structure that forms a special pattern in the central region of the fingerprint
and (ii) Minutiae details associated with the local ridge and furrow structure. In a
traditional biometric recognition system, the biometric template is usually stored on a
central server during enrollment. The candidate biometric template captured by the
biometric device is sent to the server where the processing and matching steps are
performed. This paper presents an approach to speed up the matching process by
classifying the fingerprint pattern into different groups at the time of enrollment, and
improves fingerprint matching while matching the input template with stored template.
To solve the problem, we take several aspects into consideration like classification of
fingerprint, singular points. The algorithm result indicates that this approach manages to
speed up the matching effectively, and therefore prove to be suitable for large database
like forensic divisions.

Keywords: Biometrics, identification, verification, minutiae points, singular points.

1. INTRODUCTION

Fingerprint is one of the most mature biometric traits and considered legitimate proof of evidence in
courts of law all over worldwide. Fingerprints are, therefore, used in forensic divisions worldwide for
criminal investigations. More recently, an increasing number of civilian and commercial applications are
either using or actively considering using fingerprint-based identification because of a better

International Journals of Biometric and Bioinformatics, Volume (3) : Issue (1) 1


understanding of fingerprints as well as demonstrated matching performance than any other existing
biometric technology. Modern fingerprint matching techniques were initiated in the late 16th century [1].
Henry Fauld, in 1880, first scientifically suggested the individuality and uniqueness of fingerprints. At the
same time, Herschel asserted that he had practiced fingerprint identification for about 20 years [2]. This
discovery established the foundation of modern fingerprint identification. In the late 19th century, Sir
Francis Galton conducted an extensive study of fingerprints [2]. He introduced the minutiae features for
single fingerprint classification in 1888. The discovery of uniqueness of fingerprints caused an immediate
decline in the prevalent use of anthropometric methods of identification and led to the adoption of
fingerprints as a more efficient method of identification.

FIGURE 1: Fingerprints classification involving six categories: (a) arch, (b) tented arch, (c) right loop, (d) left loop, (e)
whorl, and (f) twin loop. Critical points in a fingerprint, called core and delta, are marked as squares and triangles.
Note that an arch does not have a delta or a core.

An important advance in fingerprint identification was made in 1899 by Edward Henry, who established
the famous “Henry system” of fingerprint classification [1, 2]: an elaborate method of indexing fingerprints
very much tuned to facilitating the human experts performing (manual) fingerprint identification. In the
early 20th century, fingerprint identification was formally accepted as a valid personal identification
method by law enforcement agencies and became a standard procedure in forensics [2]. Fingerprint
identification agencies were setup worldwide and criminal fingerprint databases were established [2].
Loop, whorl, and twin loop.

1.1 Fingerprint Feature Extraction

The human fingerprint is comprised of various types of ridge patterns, traditionally classified according to
the decades-old Henry system: left loop, right loop, arch, whorl, and tented arch.



FIGURE 2: Minutiae- points on a fingerprint.

Loops make up nearly 2/3 of all fingerprints, whorls are nearly 1/3, and perhaps 5-10% are arches [3].
These classifications are relevant in many large-scale forensic applications, but are rarely used in
biometric authentication. Many types of minutiae exist, including dots (very small ridges), islands (ridges
slightly longer than dots, occupying a middle space between two temporarily divergent ridges), ponds or
lakes (empty spaces between two temporarily divergent ridges), spurs (a notch protruding from a ridge),
bridges (small ridges joining two longer adjacent ridges), and crossovers (two ridges which cross each
other) [4].

1.2 Fingerprint-Matching Process

Fingerprint matching techniques can be placed into two categories: minutiae-based and correlation
based. But the commonly used technique with minimum FAR and FRR is Minutiae-based techniques. In
this process we, first find minutiae points and then map their relative placement on the finger. However,
there are some difficulties when using this approach. It is difficult to extract the minutiae points accurately
when the fingerprint is of low quality. Also this method does not take into account the global pattern of
ridges and furrows [5]. Fingerprint Verification System is a system that determines the correspondence of
an input fingerprint with a template fingerprint stored in data base. A typical block diagram of biometric
matching systems is shown in Figure 3.



FIGURE 3: Block diagram of a typical Automatic Fingerprint Verification system.

In a verification fingerprint system, the template fingerprint image is obtained in the enrollment phase.
After that verification process takes place by a inputting the sample of the user’s fingerprint at sensor.
Such input fingerprint must be processed, in the preprocessing step. The preprocessing includes image
enhancement, gray level adjust, ridge thinning, etc. After the fingerprint image has been preprocessed,
the feature extraction block extracts the relevant information that will be used for matching with the
template fingerprint [6]. Finally a verification decision is made with the results or percentages of similarity
obtained from the matching step. Section 2 describes the work in this field and the problems associated
with this field. Section 3 describes the proposed work and the efficiency of proposed work based on
experimental calculations.

2. Related Work

Figure3 above shows the process of matching for fingerprint. It is obvious that if fingerprint templates are
stored in a particular manner then it will quite increase the efficiency of biometrics device. We have visited
Madhuban Forensic Laboratory, Karnal to know which methods are used there as there are lots of
templates in database. There we see the matching process of fingerprint identification among apx.
150000 database templates. The software being used there is FACTS (Finger Analysis Criminal Tracing
System) developed by CMC and based on the theory of Dr. Henry Faulds. This approach involves the
print of all the fingers of both hands and assign weights to each of fingers print pattern [7].

2.1 Explanation of the Henry Classification System

The Henry Classification System allows for logical categorization of ten-print fingerprint records into
primary groupings based on fingerprint pattern types. This system reduces the effort necessary to search
large numbers of fingerprint records by classifying fingerprint records according to gross physiological
characteristics.



FIGURE 4: Both palm print of a single person.

Subsequent searches (manual or automated) utilizing granular characteristics such as minutiae are
greatly simplified. Henry Classification System assigns each finger a number according to the order in
which is it located in the hand, beginning with the right thumb as number 1 and ending with the left
pinky/little as number 10. The system also assigns a numerical value to fingers that contain a whorl
pattern; fingers 1 and 2 each have a value of 16, fingers 3 and 4 have a value of 8, and so on, with the
final two fingers having a value of 1. Fingers with a non-whorl pattern, such as an arch or loop pattern,
have a value of zero. Images of various fingerprint patterns are shown already in figure1. In accordance
to the Henry Classification System, finger numbers and finger values are assigned as following: The
fingerprint record’s primary grouping is determined by calculating the ratio of one plus the sum of the
values of the whorl-patterned, even-numbered fingers; divided by one plus the sum of the values of the
whorl-patterned, odd-numbered fingers (Harling 1996). This formula is represented below [8]:

Henry Classification System Formula:

Primary Grouping Ratio (PGR) =

1+ (Sum of whorled, EVEN finger value)

1+ (Sum of whorled, ODD finger value)

If, for example, an individual has a fingerprint record with a LWAALALWLA pattern series (the series
begins with Finger 1, the right thumb and ending with Finger 10, the left pinky), the corresponding
classification ratio would be 19:1. This example is calculated below [8] :



Therefore, this individual belongs to the 19:1 primary group. If, for example, an individual does not have
any whorl-patterned fingerprints, his or her classification ratio, or primary group, would be 1:1. If an
individual has all ten fingerprints containing a whorl pattern, his or her classification ratio would be 31:31.
The Henry Classification System allows for up to 1,024 primary groupings.

2.2 Problems associated with Existing System
Above method work very efficiently when we have palm prints of all fingers of both hands. We assign
weights to the person prints and calculate PGR. On the basis on PGR factor the search goes to particular
domain and identified the proper match. But if we have only one finger print as input print, then there will
be problem as in this case we can’t find PGR factor. Further the problem can also arise if the criminal is
made some trick while giving its input prints to the system. He can change the order of his fingerprint
while giving input print, if this happen then his print can’t be matched anywhere in the system.

3. PROPOSED WORK
Proposed work is based on the theory of fingerprint classification, we store only single finger print of
person in the database. This single print can be thumb print or print of index finger. One obvious
advantage of this approach is that it will considerably reduce the amount of memory required to store the
fingerprint template as only one print is stored instead of 10 prints for an individual. Now let us see how
the proposed system will work. First let us talk about the enrollment process; in conventional system the
database contains the fingerprint templates in an ordinary manner. But here in our proposed system the
database contains the different set of templates according to classification. During the enrollment
process, sensor senses the fingerprint, then next step is feature extraction, here minutiae points are
extracted. After this step we put a classifier to check the classification of input template that whether it is
left-loop, right-loop, arch or whorl. The detail function of classifier is shown in figure 6. After classification
the input template will be stored in particular domain. A domain in the database contains the templates of
same classification. Normally the fingerprints are classified as Whorl, arch and loop. Loops make up
nearly 65% of all fingerprints, whorls are nearly 30%, and perhaps 5% are arches [3]. These
classifications are relevant in many large-scale forensic applications, but are rarely used in biometric
authentication. Since the loops are 65%, we further divide this domain into two parts i.e. left loop 32%
apx. and right loop 33% apx . So we have four different domains i) Left-Loop ii) Right-Loop iii) Arch and
iv) Whorl as shown in figure5. Now let us come to the verification process, here the finger or finger print is
placed at sensor and then its features are extracted and a final template is generated for matching. Now
this template will not matched with every templates in the database rather it extracts its classified domain
out of 4-domain and will perform match from this extracted domain. This process, no doubt will be fast
and more efficient especially when the stored database is very large that is more than 1, 00000
templates. Let D and T be the representation of the Database Template and Stored Template
respectively. Each minutia may be described by a number of attributes, including its location in the
fingerprint image, orientation, type etc. Most common minutiae matching algorithms consider each
minutiae as a triplet m= {x,y,θ}that indicates the minutiae location coordinates and the minutiae angle θ.



D= {m1,m2,…….mn} mi = {xi,yi,θi} i= 1….m
T= { m’1,m’2,…….m’n } mj = {x’j,y’j,θ’j} j= 1….n
Where m and n denotes the number of minutiae in D and T respectively.
Database Template and Stored Template and stored template will be matched, If we calculate Spatial
Distance (SD) and direction difference (DD) that will not be below than specified value r0 and θ0 or we
can write as [9].
SD (m’1, m1) = sqrt [(x’i - xi )2 + (y’i - yi )2 ] >= r0 ----------------(1)
Similarly
DD (m’1, m1) >= θ0 ----------------(2)

FIGURE 5: Proposed scheme for Fingerprint Identification

3.1 Fingerprint Classifier
Fingerprint classifiers classify the input fingerprint into four major categories namely Left-Loop, Right-
Loop, Whorl and Arch. The proposed classifiers works on the basis of singular point (Delta) extracted. If
there are two deltas then it will be counted as whorl or twin loop. If there is no delta then it will be counted
at arch. If only one delta is there then it will be either left loop or right loop.

Whorl Arch Right-Loop Left-Loop
Figures 6. Position and numbers of deltas in different finger prints.
We further find the category of loop by measuring Relative position (R). If relative position, R of delta with
respect to symmetry axis is R = 1 means the delta is on the right side of symmetry axis then it will be left
loop otherwise it will be right loop [10]. On the basis of above idea, a flowchart (figure 6) for algorithm is
designed to find the fingerprint classification.



FIGURE 6: Proposed working of Fingerprint classifier.

3.3 Performance Estimation of Proposed Scheme
Let us take the example of Madhuban forensic laboratory, Karnal where database of more than 1, 50000
templates are stored. We had performed an experiment at Madhubhan by inputting a single template at
the sensor and started to identify it from their database. The process takes 25-30 minutes to identify and
also gives 34 matched templates (equations 1, 2 satisfy for 34 templates) [7]. These 34 templates again
have to match manually and consume around 5-6 hours i.e. it’s a quite time consuming and complex task.
First let us see why the system takes so much time.
Let’s assume that time taken to perform a single match = 1 ms (1 milli seconds)

Performance of Existing System
For Best case i.e. the template is First match, Time required = 1 X 1 = 1 ms
For worst case i.e. the template is last match, Time required = 1 X 1, 50000 = 150 sec. = 25 min.
For an Average case, Time required= apx 10-20 min.

Performance of Proposed System
For Best case i.e. the template is First match, Time required = 1 X 1 = 1 ms
Now let us calculate for worst case
We have assumed 1, 50000 templates, According to classification there will be
45000 whorls (30%) + 48000 Left Loop (32%) + 49500 Right Loop (33%) + 7500 Arch (5%)
At First stage we get the template classification and accordingly particular domain will be extracted. Now
we calculate the time taken for each classification
For Whorl = 1ms X 45000 = 45 sec.
For LL = 1ms X 48000 = 48 sec.
For RL = 1ms X 49500= 49.5 sec
For Arch = 1msX 7500= 7.5 sec.



Average time = 150/4= 37.4 sec.
For an Average case, Time required= apx 20-24 sec.

Performance Factor = Time take in worst case of existing system = 25 min = 40.1
Time take in worst case of proposed system 37.4 Sec

i.e. the new approach is apx 40 time better than the existing one.

4. CONCLUSION AND FUTURE SCOPE
An approach has been presented for fingerprint matching in an effective way to reduce time. The
proposed scheme has been calculated as 40 times efficient as compared with an existing system. We
have taken four major classifications of fingerprint, whorl, arch, left-loop and right-loop. But still there is
some problem in existing system that is in our proposed system our pre-request is concern about our
database. Our database should contain only one print of every individual not both palm prints. Proposed
system will not work if we try to match fingerprint of middle finger or little finger. For that purpose we have
to store all the fingerprints of a person but the turnaround time can be reduced by study further in this
direction as 20-25 minutes are quite lengthy time to wait and moreover there is not single match in the
result rather there are 35-50 templates matched.

5. REFERENCE:
1. D. Maltoni, D. Maio, A. K. Jain, S. Prabhakar. “Handbook of Fingerprint Recognition”. Springer- Verlag,
2003.

2. H. C. Lee and R. E. Gaensslen, Advances in Fingerprint Technology, Elsevier, New York, 1991.
3. American university of beirut faculty of engineering and architecture department of electrical and
computer engineering eece695c – adaptive filtering and neural networks fingerprint identification–project
2
4. Biometrics Information Group www.biometricsinfo.org.
5. A. K. Jain, R. Bolle, S. Pankanti (eds), Biometrics: Personal Identification in Networked Society, Kluwer
Academic, December 1998.
6. A. K. Hrechak and J. A. McHugh, Automated Fingerprint Recognition using Structural Matching,
Pattern Recognition, Vol. 23, No. 8, 1990.
7. Manuals, Forensic Science Laboratory, Madhuban, Karnal
8. The Henry Classification System Copyright © 2003 International Biometric Group
9. L. Hong, Automatic Personal Identification Using Fingerprints, PhD Thesis, Michigan State University,
1998
10. Anil Jain Sharath Pankanti, Fingerprint Classification and Matching 2004.


D.Shanthi, Dr.G.Sahoo & Dr.N.Saravanan

Designing an Artificial Neural Network Model for the Prediction of
Thrombo-embolic Stroke

D.Shanthi dshan71@gmail.com
Department of Computer Science and Engineering
Birla Institute of Technology
Budaiya, P.O.Box 31320, Kingdom of Bahrain

Dr.G.Sahoo gsahoo@bitmesra.ac.in
Department of Information Technology
Mesra, Ranchi, India

Dr.N.Saravanan saranshanu@gmail.com
Department of Computer Science and Engineering
Budaiya, P.O.Box 31320, Kingdom of Bahrain

ABSTRACT

In this study, a functional model of ANN is proposed to aid existing diagnosis
methods. This work investigated the use of Artificial Neural Networks (ANN) in
predicting the Thrombo-embolic stroke disease. The Backpropogation algorithm
was used to train the ANN architecture and the same has been tested for the
various categories of stroke disease. This research work demonstrates that the
ANN based prediction of stroke disease improves the diagnosis accuracy with
higher consistency. This ANN exhibits good performance in the prediction of
stroke disease in general.

Keywords: Artificial Intelligence, BPN, Neural Network, Thrombo-embolic Stroke.

1. INTRODUCTION
Neural networks provide a very general way of approaching problems. When the output of the
network is continuous, it is performing prediction and when the output has discrete values, then it
is doing classification. A simple rearrangement of the neurons and the network becomes adept at
detecting clusters. Computer Assisted Decision Support in medicine has at least the role of
enhancing the consistency of care. Secondly, it has the potential to cover rare conditions, since
no clinical expert can be expected to possess encyclopedic knowledge of all of the exceptional
manifestations of diseases, even within a specialist domain. Thirdly, the expanding range of
patient information that is made available in electronic form, makes it feasible to more accurately
quantify important clinical indicators, such as the relative likelihood for competing diagnoses or
the clinical outcome. In some cases, computer-assisted diagnoses have been claimed to be
even more accurate than those by clinicians.

Stroke is a life-threatening event in which part of the brain is not getting enough oxygen. There
are different types of stroke namely Brain Attack, Embolic Stroke, Thrombotic Stroke, Ischemic
Stroke, Cerebrovascular Accident (CVA). Medical personnel treating a stroke are challenged to

International Journals of Biometric and Bioinformatics (IJBB), Volume (3) : Issue (1) 10


treat the patient as quickly as possible to avoid permanent tissue damage or death. Strokes were
responsible for more deaths and nearly half of those deaths occurred outside of a hospital.
Stroke is the third leading cause of death, behind heart disease and cancer. Most recovery
occurs during the first few months following a stroke. According to the National Institute of Health,
the risk of stroke is greater – and the recovery process is slower. Thrombo embolic strokes are
caused by fatty deposits (plaques) that have built up in the arteries carrying blood to the brain.
This slows blood flow and can cause clots to form on the plaques that narrow or block the flow of
oxygen and nutrients to the brain. It is also caused by a blood clot formed in another part of the
body that breaks loose, travels through the bloodstream, and blocks an artery carrying oxygen
and nutrients to the brain. When travelling through the body the blood clot is called an embolus
[1]. A hemorrhagic stroke is caused when an artery supplying blood bleeds into the brain. The
broken blood vessel prevents needed oxygen and nutrients from reaching brain cells. One type of
hemorrhagic stroke is caused when an artery that has weakened over time bulges (called an
aneurysm) and suddenly bursts [2]. Thrombo-embolic Stroke can be classified as Transient
Ischemic attacks (TIA), Evolving Stroke, Completed Stroke, Residual Squeal, Classical Stroke,
Inappropriate Stroke, Anterior Cerebral Territory Stroke, Posterior Cerebral Stroke, Middle
Cerebral Territory Stroke. Hemorrhagic Stroke can be classified as Cerebellar stroke, Thalamic
Stroke and Cortical Stroke. In this paper, we propose an artificial neural network model for the
prediction of Thrombo Embolic Stroke disease. The rest of the paper discusses about related
studies, the proposed model, results and discussion along with conclusion.

2. RELATED STUDIES OF ANN IN MEDICINE
ANNs appear to be a valid candidate for the reliability analysis. Given a number of predictor
variables, an opportunely structured and trained Multi Layer Perceptron (MLP) can identify the
“causal path” leading to a certain value of the potential objective variables, with a certain degree
of confidence. [3]. Several studies have applied neural networks in the diagnosis of
cardiovascular disease, primarily in the detection and classification of at-risk people from their
ECG waveforms [4]. In the works of [5], the application of neural networks to classify normal and
abnormal (pathological) ECG waveforms and the abnormal ECG recordings had six different
disease conditions. The classifier was able to recognize these waveforms with 70.9% accuracy.

The Study [6] suggested that the role of the ANN, which uses non-linear statistics for pattern
recognition in predicting one-year liver disease-related mortality using information available
during initial clinical evaluation. MLP with sigmoidal feed-forward and standard Back-Propagation
(BP) learning algorithm was employed as a forecaster for bacteria-antibiotic interactions of
infectious diseases. Comparing ANN ensembles with logistic regression models we found the
former approach to be better in terms of ROC area and calibration assessments. Both ANN and
logistic regression models showed intra-method variations, as a result of training the models with
different parts of the study population. This variation was larger for the ANN ensemble models [7].

The studies of application of ARTMAP in medicine include classification of cardiac arrhythmias [8]
and treatment selection for schizophrenic and unipolar depressed in-patients [9]. Another study
revealed that fully connected feed forward MLP and BP learning rule, were able to predict
patients with colorectal cancer more accurately than clinicopathological methods. They indicate
that NN predict the patients’ survival and death very well compared to the surgeons. The study
[10] presents their study for the diagnosis of Acute Myocardial Infarction. The results show that
NN performance is 0.84 and 0.85 under ROC.

A neural network can provide a considerable improvement in the diagnosis of early acute allograft
rejection, though further development work is needed before this becomes a routine diagnostic
tool. The selection of cases used to train the network is crucial to the quality of its performance.
There is scope to improve the system further by incorporating clinical information[11]. Another
methodology, which is based on ANNs, has been developed for the detection of ischaemic
episodes in long-duration ECG recordings [12] . The raw ECG signal containing the ST segment



and the T wave of each beat is the input to the beat-classification system, and the output is the
classification of the beat. The use of the ANN model as a data mining tool is very promising for
new knowledge discovery in nephrology, to model complex behaviour of different molecular
markers of dialysis treatment and for online treatment monitoring. [13]. The Study [14],
presented a fully automated method using ANNs were compared with the clinical interpretation.
The neural networks trained with both perfusion and ECG-gated images had a 4–7% higher
specificity compared with the corresponding networks using perfusion data only, in four of five
segments compared at the same level of sensitivity. The addition of functional information from
ECG-gated MPS is of value for the diagnosis of myocardial infarction using an automated method
of interpreting myocardial perfusion images.

3. THE PROPOSED MODEL
3.1 Patient Data
The data for this study have been collected from 50 patients who have symptoms of stroke
disease. The data have been standardized so as to be error free in nature. All the fifty cases are
analyzed after careful scrutiny with the help of the Physicians. Table-1 below shows the various
input parameters for the prediction of stroke disease.

Sl.No. Parameters
1 Age
2 Sex
3 Pre-stroke mobility
4 Hypertension
5 Diabetes Mellitus
6 Myocardial infarction
7 Cardiac failure
8 Atrial fibrillation
9 Smoking
10 High blood cholesterol
11 Alcohol abuse
12 Weakness of Left Arm and Left leg
13 Weakness of Right Arm and Right leg
14 Slurring of Speech
15 Giddiness
16 Headache
17 Vomiting
18 Memory Deficits
19 Swallowing Difficulties
20 Loss of Vision
21 Isolated vertigo
22 Transient Double Vision
23 Sudden difficulty in walking , dizziness or loss
of balance
24 Hand / Leg numbness
25 Transient loss of consciousness

TABLE 1: Input Parameters for Prediction of Stroke
3.2 Feature Selection
Data are analyzed in the dataset to define column parameters and data anomalies. Data analysis
information needed for correct data preprocessing. After data analysis, the values have been
identified as missing, wrong type values or outliers and which columns were rejected as
unconvertible for use with the neural network[15]. Feature selection methods are used to identify
input columns that are not useful and do not contribute significantly to the performance of neural
network. In this study, Backward stepwise method is used for input feature selection. The
removal of insignificant inputs will improve the generalization performance of a neural network.
This method begins with all inputs and it works by removing one input at each step. At each step,
the algorithm finds an input that least deteriorates the network performance and becomes the



candidate for removal from the input set. Table 2 shows the finalized input parameters after
applying feature selection method.

Input column name Code Importance %
Hypertensive X1 1.626103
Diabetes X2 4.229285
Myocardial X3 0.043249
Cardiac failure X4 0.001659
Atrial fibrillation X5 0.034991
Smoking X6 2.061142
Blood cholesterol X7 8.831646
left arm&leg X8 19.209636
Right arm &leg X9 2.832501
Slurring X10 1.776497
Giddiness X11 3.64891
Headache X12 15.646755
Vomiting X13 0.535259
memory deficits X14 1.485701
Swallowing X15 5.224413
Vision X16 7.366008
Double vision X17 1.13867
Vertigo X18 15.471004
Numbness X19 0.136173
Dizziness X20 8.700397

TABLE 2: Percentage of Importance of Input Data

3.3 Neural Network Architecture
The architecture of the neural network used in this study is the multilayered feed-forward network
architecture with 20 input nodes, 10 hidden nodes, and 10 output nodes. The number of input
nodes are determined by the finalized data; the number of hidden nodes are determined through
trial and error; and the number of output nodes are represented as a range showing the disease
classification. The most widely used neural-network learning method is the BP algorithm [16].
Learning in a neural network involves modifying the weights and biases of the network in order to
minimize a cost function. The cost function always includes an error term a measure of how
close the network's predictions are to the class labels for the examples in the training set.
Additionally, it may include a complexity term that reacts a prior distribution over the values that
the parameters can take.

The activation function considered for each node in the network is the binary sigmoidal function
-x
defined (with σ = 1) as output = 1/(1+e ), where x is the sum of the weighted inputs to that
particular node. This is a common function used in many BPN. This function limits the output of
all nodes in the network to be between 0 and 1. Note all neural networks are basically trained
until the error for each training iteration stopped decreasing. Figure 1 shows the architecture of
the specialized network for the prediction of stroke disease. The complete set of final data (20
inputs) are presented to the generic network, in which the final diagnosis corresponds to output
units.



FIGURE 1: Artificial Neural Network Architecture for the prediction of stroke disease

The net inputs and outputs of the j hidden layer neurons can be calculated as follows
N +1
net h = ∑ W ji xi
j
t =1

y j = f (net h )
j
Calculate the net inputs and outputs of the k output layer neurons are
J +1
netko = ∑ Vkj y j
j =1

Z k = f (netko )
Update the weights in the output layer (for all k, j pairs)
vkj ← vkj + cλ ( d k − Z k ) Z k (1 − Z k ) y j
Update the weights in the hidden layer (for all i, j pairs)



k
w ji ← w ji + cλ 2 y j (1 − y j ) xi (∑ (d k − zk ) zk (1 − Z k )vkj )
k =1
Update the error term
k
E ← E + ∑ (d k − zk )2
k =1
and repeat from Step 1 until all input patterns have been presented (one epoch). If E is below
some predefined tolerance level , then Stop. Otherwise, reset E = 0, and repeat from Step 1 for
another epoch.

The inputs to the models were 20 variable training parameters and the output indicated the
point at which training should stop. The following are the results generated from the input given
to the neural network after going through the process of careful training, validation and testing
using Neuro Intelligence tool. Table 3 shows the various categories of Stroke diseases and their
classification.

Output Code
TIA D1
Left Hemiplegia D2
Right Hemiplegia D3
Dysphasia D4
Monoplegia D5
Left Hemianopia D6
Aphasia D7
Right Hemianesthesia D8
Dysphagia D9
Quadruplegia D10

TABLE 3: Output Classification

4. RESULTS AND DISCUSSION
The Data have been analyzed using Neuro-intelligence tool [17].During analysis, the
column type is identified. During data analysis, the last column is considered as the target one
and other columns will be considered as input columns. The dataset is divided in to training set,
validation set and test set.

Sl.No Data Partition set Records Percentage
1. Training set 34 68%
2. Validation set 8 16%
3. Test set 8 16%
4. Ignored set 0 0%
Total 50 100%

TABLE 4: Data Partition Set

Training a neural network is the process of setting the best weights on the inputs of each of the
units. The goal is to use the training set to produce weights where the output of the network is as
close to the desired output as possible for as many of the examples in the training set as
possible. Also it has been proved that Genetic Algorithm and Back-Propagation neural network
hybrids in selecting the input features for the neural network reveals the performance of ANN can
be improved by selecting good combination of input variables [18]. The training set is a part of
the input dataset used for neural network training, i.e. for adjustment of network weights. The
validation set is a part of the data are used to tune network topology or network parameters other
than weights. For example, it is used to define the number of units of to detect the moment when
the neural network performance started hidden to deteriorate. To choose the best network (i.e. by
changing the number of units in the hidden layer) the validation set is used. The test set is a part



of the input data set used to test how well the neural network will perform on new data. The test
set is used after the network is ready (trained), to test what errors will occur during future network
application. This set is not used during training and thus can be considered as consisting of new
data entered by the user for the neural network application.

Figure 2 shows the various data set errors with respect to training set, validation set and the best
network. After training through repeated iterations it reaches the level of best network.

FIGURE 2: Data Set Errors

In the diagnosis of stroke , it is not always possible to make a clear-cut determination of disease,
because of variability in the diagnostic criteria, age at onset, and differential presentation of
disease. Mapping such diseases is greatly simplified if the data present a homogeneous genetic
trait and if disease status can be reliably determined. Here, we present an approach to
determination of disease status, using methods of artificial neural-network analysis. The Network
errors have been shown graphically in figure 3. After 150 iterations, the network error has been
decreased and from 300 iterations it is almost 0.

FIGURE 3: Network Errors vs. Training Error

The trained network has been tested with a test set, in which the outcomes are known but not
provided to the network, to see how well the training has worked.. We used diagnostic criteria and
disease status to train a neural network to classify individuals as "affected" by several categories
of stroke as given below .

The analysis shows clearly that 32% of the respondents have the symptoms of Left Hemiplegia;
14% each have the symptoms of TIA and Right Hemiplegia respectively; 10% of the patients
have the symptoms of Dysphasia and 6% are suffering from Monoplegia. 8% each have the
symptoms of Left Hemianopia and Aphasia. In the meantime, 4% have the symptoms of Right
Hemianesthesia. 2% each have the symptoms of Dysphagia and Quadruplegia respectively and
is diagrammatically depicted in figure 4.



Disease Classification

40
No of Cases(%)

30

20 %
10
0
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10

% 14 32 14 10 6 8 8 4 2 2
Diseases

FIGURE 4: Various Stroke Diseases vs. Number of Cases

5. CONCLUSION
Neural networks have been proposed as useful tools in decision making in a variety of medical
applications. Neural networks will never replace human experts but they can help in screening
and can be used by experts to double-check their diagnosis. In general, results of disease
classification or prediction task are true only with a certain probability. This work described here
shows that the prediction of risk from stroke gives best results on the dataset used. The results
generated by this system have been verified with the physicians and are found correct. This ANN
model helps the doctors to plan for a better medication and provide the patient with early
diagnosis as it performs reasonably well even without retraining. In conclusion, when the ANN
was trained and tested after optimizing the input parameters, the overall predictive accuracy
obtained was 89%.

6. REFERENCES
1. Mohr, J.P., (2001). Stroke Analysis, 4th Edition, Oxford Press, New York.

2. American Heart Association. Heart Disease and Stroke Statistics — 2004 Update.
Dallas, Tex.: American Heart Association; 2003.

3. Maurizio Bevilacqua (2005), Failure rate prediction with artificial neural networks, Journal
of Quality in Maintenance Engineering Vol. 11 No. 3, 2005 pp. 279-294 Emerald Group
Publishing Limited 1355-2511

4. Nazeran, H., & Behbehani, K. (2001). Neural networks in processing and analysis of
biomedical signals. In M. Akay (Ed.), Nonlinear biomedical signal processing: Fuzzy
logic, neural networks and new algorithms, pp. 69–97.

5. Celler, B.G., & Chazal, P. (1998). Low computational cost classifiers for ECG diagnosis
using neural networks. Proceedings of the International Conference of Engineering in
Medicine & Biology Society (EMBC 1998) ,pp. 1337–1340.

6. Rupa banerjee et al.,(2003) , Predicting mortality in patients with cirrhosis of liver with
application of neural network technology, 2003 Blackwell Publishing Asia Pvt. Ltd.



7. Green, M. et al.(2006).Comparison between neural networks and multiple logistic
regression to predict acute coronary syndrome in the emergency room , Elsevier B.V.

8. Ham, P.M. and Han, S. "Classification of cardiac arrhythmias using fuzzy artmap," IEEE
Transactions on Biomedical Engineering, 43(4): 425–430 (1996).

9. Modai, I., Israel, A., Mendel, S., Hines, E.L. and Weizman, R., "Neural network based on
adaptive resonance theory as compared to experts in suggesting treatment for
schizophrenic and unipolar depressed in-patients," Journal of Medical Systems, 20(6):
403–412 (1996).

10. Mattias Ohlsson et al., (2006). The added value of ECG-gating for the diagnosis of
myocardial infarction using myocardial perfusion scintigraphy and artificial neural
networks, Clinical Physiological Functional Imaging 26, pp301–304.

11. Furness P N, Levesley J, Luo Z, Taub N, Kazi J I, Bates W D & Nicholson M L (1999) , A
neural network approach to the biopsy diagnosis of early acute renal transplant rejection
Histopathology 35, 461–467.

12. Papaloukas, C.,Fotiadis, D. I.,Likas, A., &Michalis, L. K. (2002b). An ischemia detection
method based on artificial neural networks. Artificial Intelligence in Medicine, 24, 167–
178.

13. Elmer Andres Fernández, Rodolfo Valtuille, Jesus Rodriguez Presedo, and Peter
Willshaw (2005) Comparison of Standard and Artificial Neural Network Estimators of
Hemodialysis Adequacy , International Center for Artificial Organs and Transplantation.
29(2):159–165, Blackwell Publishing, Inc.

14. Peter Gjertsson et al.,(2006) The added value of ECG-gating for the diagnosis of
myocardial infarction using myocardial perfusion scintigraphy and artificial neural
networks, Clinical Physiological Functional Imaging 26, pp301–304.

15. Baxt WG. Application of artificial neural networks to clinical medicine. Lancet.
1995; 346 (8983): 1135- 8.

16. Rumelhart, D.E., McClelland, J.L., and the PDF Research Group (1986), Parallel
Distributed Processing, MA: MIT Press, Cambridge.

17. Neuro Intelligence using Alyuda. Source Available at www.alyuda.com. Last Accessed
10 May 2008.

18. D.Shanthi, G.Sahoo & N.Saravanan (2008). Input Feature Selection using Hybrid Neuro-
Genetic Approach in the diagnosis of Stroke. International Journal of Computer Science
and Network Security, ISSN: 1738-7906, Vol.8, No.12, pp.99-107.


International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue (1)

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue (1) (20)

International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue (1)