You are on page 1of 10

Pattern Recognition for Diagnostic of Pathologies on the Vertebral

Column
Thibaut Judicael BAH
Supervisor: Prof. Bekir Karlik
Department of Computer Engineering
Selçuk University, Konya, Turkey

ABSTRACT

Accurate and computationally efficient means to predict the Diagnostic of pathologies on the
vertebral column has been the subject of considerable research effort in recent years. The
vertebral column is a complicated anatomical structure which is composed of the intervertebral
discs and the vertebrae. Both components develop special degenerative changes and morphologic
features during life. There are some vertebral column pathologies like herniation and
spondylolisthesis which sometimes can press on a nerve inside the spine and causes radical pain
spreading to other parts of the body. The medical diagnosis is critical to decide methods of
treatment [1]. Radiologists and doctors look at the images taken by medical imaging devices in
order to diagnosis. However, due to personal errors such as fatigue, inexperience, oversight,
wrong diagnosis is possible. In order to prevent these, computer-aided diagnostic (CAD)
methods are mostly preferred [2]. This paper present an automatic diagnosis of vertebral column
pathologies, Lumbar Disc Hernia and Spondylolisthesis.

Keywords: Lumbar Disc Hernia, Spondylolisthesis, Artificial Neural Network, Multilayer Layer
Perceptron (MLP)

I. Introduction

Over the last decade, we have been assisting to an increasing number of Machine Learning
(ML) techniques, such as Support Vector Machines (SVM), Artificial Neural Networks (ANN)
or Naïve Bayes, applied to several fields [3].
In fact, Pattern Recognition has been commonly used with rapidly increasing success as
in medicine, science or commerce. Especially in medicine, where extensive amount of data and
knowledge stored in clinical databases, main contributing factor influencing the development of
such systems has been the additional demands for a more powerful, flexible and transparent
technique [1].
The Vertebral Column

The vertebral column is a system composed by a group of vertebras, invertebrate discs,


nerves, muscles, medulla and joints. The main functions of the vertebral column are as follow: (i)
human body support axle; (ii) osseous protector of the spine medulla and nervous roots; and (iii)
body's movement axles, making movement possible in three levels: frontal, sagittal and
transversal. The human vertebral column usually consists of 33 vertebrae; the upper 24
are articulating vertebrae, separate by intervertebral discs.

Figure1: Human vertebral column

This complex system can suffer dysfunctions that cause backaches with very different
intensities. Disc hernia and spondylolisthesis are examples of pathologies of the vertebral
column that cause intense pain. They result of small or several traumas in the column that
gradually injures the structure of the intervertebral disc [3].

A herniated disc occurs when the inner core leaks out through a damaged part of the outer
portion of the disc. A herniated disc can occur in any area of the spine, including the lumbar
(lower back), thoracic (upper back) and cervical (neck) regions.

Fig.2: Herniated disc


Spondylolisthesis is the forward displacement of a vertebra, especially the fourth and the fifth
lumbar vertebra, most commonly occurring after a break or fracture. Backward displacement is
referred to as retrolisthesis.

Fig.3: Spondylolisthesis

II. Related Work

B. Karlik et. al. proposed a diagnosis of lumbar disc hernia using wavelet transform and
neural network [1], A. R. Rocha Neto and G. A. Barreto made a work about the application of
ensembles of classifiers to the diagnosis of pathologies of the vertebral column: a comparative
analysis [4], R.S. Alomari et al. proposed a Labelling of Lumbar Discs Using Both Pixel- and
Object-Level Features With a Two-Level Probabilistic Model [5], Da Rocha Neto et al. made a
research on diagnostic of pathology on the vertebral column with embedded reject option [6].

III. Materials and Methods

A. Data set
Vertebral column data set used in this paper is taken from UCI Machine Learning Repository.
This Biomedical data set was built by Dr. Henrique da Mota during a medical residence period in
the Group of Applied Research in Orthopedics (GARO) of the Centre Médico-Chirurgical de
Réadaptation des Massues, Lyon, France. The data have been organized in order to classify
patients as belonging to one out of three categories: Normal (100 patients), Disk Hernia (60
patients) or Spondylolisthesis (150 patients).
Each patient is represented in the data set by six biomechanical attributes derived from the shape
and orientation of the pelvis and lumbar spine (in this order): pelvic incidence, pelvic tilt, lumbar
lordosis angle, sacral slope, pelvic radius and grade of spondylolisthesis. The following
convention is used for the class labels: DH (Disk Hernia), Spondylolisthesis (SL) and Normal
(NO).

B. Artificial Neural Network Approach

Artificial Neural systems are physical cellular systems which can acquire, store and utilize
experimental knowledge. The distinguished characteristics of neural networks have played an
important role in a wide variety of applications. Powerful learning algorithms and self-
organizing rules allow ANN to self-adapt as per the requirements in a continually changing
environment (adaptability property) [7]. The ability to perform tasks involving nonlinear
relationships and pathology diagnosis make ANN a good candidate for classification and
prediction (nonlinear processing property).

Commonly neural networks are adjusted, or trained, so that a particular input leads to a
specific target output. Typically many such input/target pairs are needed to train a network.
Neural network have been trained to perform complex functions in in various fields, including
pattern recognition, identification, classification, speech, vision, and control systems. Today
neural networks can be trained to solve problems that are difficult for conventional computers or
human beings [8].

C. Multi-layer Perceptron (MLP) Architecture for Backpropagation Algorithm

The Multi-layer perceptron is the most widely used type of neural network. It is both simple and
based on solid mathematical grounds. Input quantities are processed through successive layers of
“neurons”. There is always an input layer, with a number of neurons equal to the number of
variables of the problem; and an output layer, where the perceptron response is made available,
with a number of neurons equal to the desired number of quantities computed from the inputs.
The layers in between are called

Fig.4: Used MLP structure of ANN

“Hidden” layers. With no hidden layer, the perceptron can only perform linear tasks. All
problems which can be solved by a perceptron can be solved with only one hidden layer [8].
In this study, a three-layered feed-forward (MLP) neural network architecture was utilized
and trained with the error backpropagation algorithm. The features are used as the input vectors.
The Backpropagation algorithm is described step by step as [9]:

1. Initialization: Set all the weights and biases to small real random values.
2. Presentation of input and desired outputs: Present the input vector x(1), x(2),…,x(N) and
corresponding desired response d(1),d(2),…,d(N), one pair at a time, where N is the
number of training patterns.
3. Calculation of actual outputs: Use Eq. (8) to calculate the output signals 𝑦1 , 𝑦2 , … . , 𝑦𝑁𝑀

𝑁𝑀−1
(𝑀−1) (𝑀−1) (𝑀−1)
𝑦𝑖 = 𝜑 ( ∑ 𝑤𝑖𝑗 𝑥𝑗 + 𝑏𝑖 ), 𝑖 = 1, … , 𝑁𝑀−1 (8)
𝑗=1

4. Adaptation of weights (wij) and biases (bi):


(𝑙−1)
∆𝑤𝑖𝑗𝑙−1 (𝑛) = 𝜇𝑥𝑗 (𝑛) . 𝛿𝑖 (𝑛) (9)

(𝑙−1)
∆𝑏𝑖𝑙−1 (𝑛) = 𝜇. 𝛿𝑖 (𝑛) (10)

And
(𝑙−1)
𝜑′(𝑛𝑒𝑡𝑖 ) [𝑑𝑖 − 𝑦𝑖 (𝑛)], 𝑙=𝑀
(𝑙−1)
𝛿𝑖 (𝑛) = { (𝑙−1) (𝑡) (11)
𝜑′(𝑛𝑒𝑡𝑖 ) ∑ 𝑤𝑘𝑖 . 𝛿𝑘 (𝑛), 1≤𝑙≤𝑀
𝑘

Where 𝑥𝑗 (𝑛) = output of node j at iteration n, l is layer, k is the number of output nodes of
neural network, M is output layer, φ is activation function. The learning rate is demonstrated by
μ. It may be noted here that a large value of the learning rate may lead to faster convergence but
may also result in oscillation.
With the purpose of achieving faster convergence with minimum oscillation, a momentum term
may be added to the basic weight updating equation. After completing the training procedure of
the neural network, the weights of MLP are frozen and ready for use in the testing mode.

In this study the network is trained by using Weka-3-6-10jre-x64. 88% (273 instances)
of the available data is used for training, and 12% (37 instances) for testing. The data for each
class are chosen randomly from the data set.

The configuration of the PC can be listed as: AMD A8-4500 APU with Radeon(tm) HD
Graphics 1.90 GHz, 4 GB RAM Memory, 500 GB Hdd.

In the network, we used back-propagation function. This function is a network training


function that updates weight and bias values according to optimization method. The nodes in this
network are all sigmoid. Parameters to fix before train and test the network in weka:
Fig.5: MLP parameters in weka

MLP Parameters Definitions

GUI -- Brings up a gui interface. This will allow the pausing and altering of the neural network
during training.
autoBuild -- Adds and connects up hidden layers in the network.

debug -- If set to true, classifier may output additional info to the console.

decay -- This will cause the learning rate to decrease. This will divide the starting learning rate
by the epoch number, to determine what the current learning rate should be. This may help to
stop the network from diverging from the target output, as well as improve general performance.

hiddenLayers -- This defines the hidden layers of the neural network. This is a list of positive
whole numbers. 1 for each hidden layer. Comma separated. To have no hidden layers put a
single 0 here.

learningRate -- The amount the weights are updated.

momentum -- Momentum applied to the weights during updating.


nominalToBinaryFilter -- This will preprocess the instances with the filter. This could help
improve performance if there are nominal attributes in the data.

normalizeAttributes -- This will normalize the attributes. This could help improve performance
of the network. This is not reliant on the class being numeric. This will also normalize nominal
attributes as well so that the nominal values are between -1 and 1.

normalizeNumericClass -- This will normalize the class if it's numeric. This could help improve
performance of the network, It normalizes the class to be between -1 and 1. Note that this is only
internally, the output will be scaled back to the original range.

reset -- This will allow the network to reset with a lower learning rate. If the network diverges
from the answer this will automatically reset the network with a lower learning rate and begin
training again.

seed -- Seed used to initialize the random number generator. Random numbers are used for
setting the initial weights of the connections between nodes, and also for shuffling the training
data.

TrainingTime -- The number of epochs to train through.

validationSetSize -- The percentage size of the validation set.

validationThreshold -- Used to terminate validation testing. The value here dictates how many
times in a row the validation set error can get worse before training is terminated.

Fig.7: MLP training result of ANN in weka


IV. Experimental Results

In this study, the MLP used structure is 6:6:1, which are 6 neurons in the input layer, 6
neurons in the hidden layer and only one in the output layer. Optimum neurons of hidden
layer was found as 6 after using different number of neurons (see Fig.4).

Table 1: Structure and parameter of MLP


Number of layer 3 (Input-Hidden-Output)
Number of neurons 6-6-1 (Input-Hidden-Output)
Activation function Sigmoid
Learning Algorithm Back-propagation
Output Classes Hernia – Spondylolisthesis - Normal

The architecture and the parameters of the network is given in Table 1. The network was
trained for 500 epoch with back-propagation function and show the following results as it can
be seen in Fig.6.

Fig.7: Training and Test result of MLP


Table 2: Confusion Matrix
Disc Hernia Spondylolisthesis Normal
Disc Hernia 6 0 0
Spondylolisthesis 0 22 0
Normal 1 0 6

The experimental results are given in Fig.6 and 7. The training performance of the pathology on
vertebral column diagnosing system is 94.18%, and the testing performance is 97.297%. As it
can be seen from the confusion matrix table, all of the testing instances Hernia (6 instances) and
Spondylolisthesis (22 instances) are classified correctly, and only one instance for Normal (9
instances) is misclassify. So that the classification rate for Hernia and Spondylolisthesis is 100%
and the rate for Normal is 88.9%. The training time is recorded as 1.16 seconds.

A. Performance Comparison

We have compared the proposed approach to other approaches that have previously been
published in the literature.
In Table 3, we show the classification results reported by other researchers as well as the results
of the proposed approach. All these classification results are based on the vertebral column
dataset taken from UCI Machine Learning Repository.
From Table 3, we can see that the proposed approach performs better on vertebral column
diagnosis than most of the published approaches.

Table 3: performance comparison for diagnosis accuracy of vertebral column


Algorithms Classifiers Nb of classes Accuracy (%)
A. R. Rocha Neto and G. A. Barreto [4] MLP 3 93.39
A. R. Rocha Neto and G. A. Barreto [4] C-SVM 3 95.86
A. R. da Rocha Neto et. Al [6] SVM (KMOD) 2 85.9
Our approach MLP 3 97.30

V. Conclusions

In this study, artificial neural networks based diagnosis system for pathology on vertebral
column is presented. For this purpose. The study demonstrates an application for classification of
medical dataset. Supervised learning approach and multilayer perceptron model are applied and
successful results are obtained and presented.
Model is trained with 273 samples and tested with remaining 37 samples. The feature
vectors were fed into the multi-layered ANN as input. After training the network, testing samples
were classified with an accuracy of 97.2973%. So this proposed diagnosing system can be used
for helping to make decision on vertebral column degenerative pathologies such as Disc Hernia
and Spondylolisthesis as a support tool for orthopaedists. The performance of the proposed
system can be improved with more images in the future.
References

1. B.Karlik, S.K., Diagnosis of Lumbar Disc Hernia Using Wavelet Transform and Neural Network.
Ukrainian Journal of Telemedicine and Medical Telematics, 2009. vol. 7(no.1): p. pp. 10-15.

2. Unal, Y., Kocer, H. E., & Akkurt, H. E., Automatic Diagnosis of Intervertebral Degenerative
Disk Disease Using Artificial Neural Network. 6th International Advanced Technologies
Symposium (IATS’11), 2011: p. pp. 16-18.

3. da Rocha Neto, A.R., Sousa, R., Barreto, G. D. A., & Cardoso, J. S. , Diagnostic of pathology on
the vertebral column with embedded reject option. Pattern Recognition and Image Analysis,
2011: p. pp. 588-595.

4. Neto, A.R., & Barreto, G. A., WCI 04 On the Application of Ensembles of Classifiers to the
Diagnosis of Pathologies of the Vertebral Column: A Comparative Analysis. Latin America
Transactions, IEEE (Revista IEEE America Latina), 2009. 7(4): p. 487-496.

5. Alomari, R.S., J.J. Corso, and V. Chaudhary, Labeling of Lumbar Discs Using Both Pixel- and
Object-Level Features With a Two-Level Probabilistic Model. Medical Imaging, IEEE
Transactions on, 2011. 30(1): p. 1-10.

6. da Rocha Neto, A., et al., Diagnostic of Pathology on the Vertebral Column with Embedded
Reject Option, in Pattern Recognition and Image Analysis, J. Vitrià, J. Sanches, and M.
Hernández, Editors. 2011, Springer Berlin Heidelberg. p. 588-595.

7. Karlik, B., S. Aydin, and M. Pakdemirli, Vibrations of a beam-mass systems using artificial
neural networks. Computers & structures, 1998. 69(3): p. 339-347.

8. İşeri, A. and B. Karlık, An artificial neural networks approach on automobile pricing. Expert
Systems with Applications, 2009. 36(2): p. 2155-2160.

9. Haykin, S. and N. Network, A comprehensive foundation. Neural Networks, 2004. 2(2004).

You might also like