1

Application of Artificial Neural Network in the cancer diagnosis

By: Saeid Afshar

Ph.D. student of molecular medicine

Hamedan University of Medical Sciences

Department : molecular medicine and genetics.

2

3

Artificial Neural Networks

An artificial neural network (ANN) is a computational model that attempts to account for the parallel nature of the human brain. An (ANN) is a network of highly interconnecting processing elements (neurons) operating in parallel.

4

5

6

7

Artificial Neural Networks

The simplest kind of neural network is a single-layerperceptron network, which consists of a single layer of output nodes; the inputs are fed directly to the outputs via a series of weights.

8

Artificial Neural Networks

Input 1 (x1) = 0.6Input 2 (x2) = 1.0Weight 1 (w1) = 0.5Weight 2 (w2) = 0.8

Threshold = 1.0

x1w1 + x2w2 = (0.6 x 0.5) + (1 x 0.8) = 1.1

Now we compare our input total to the

perceptron's activation threshold. In this example

the total input (1.1) is higher than the activation

threshold (1.0) so the neuron would fire

9

10

Artificial Neural NetworksTransfer Functions :

Three of the most commonly used functions are: shown below:

11

Artificial Neural NetworksMulti-layer Artificial Neural Network:

A subgroup of processing element is called a layer in the network. The first layer is the input layer and the last layer is the output layer. Between the input and output layer, there may be additional layer(s) of units, called hidden layer(s).

12

13

14

The advantages of a neural network

A neural network can perform tasks that cannot be done by a linear program.Most applications of artificial neural networks to medicine are classification problems; that is, the task is on the basis of the measured features to assign the patient to one of a small set of classes

15

The advantages of a neural network

a neural network learns and does not need to be reprogrammed.

Artificial neural networks provide a powerful tool to help physician to analyze, model and make sense of complex clinical data across a broad range of medical applications.

16

Utilization ANN in Disease Forecasting Neural network has been established of their potentials in many domains related with medical forecasting and diagnosis disease.

Although, Neural networks never replace the human experts instead they can helpful for decision making, classifying, screening

17

Utilization ANN in Disease Forecasting

Title: Method: Results:

Diabetes Mellitus Forecast Using Various Types of Artificial Neural Network

Back-propagation algorithm

The best performance was observed as 82.10% in the MLP with 8-20-1 structure.

Recognition and prediction of leukemia with Artificial Neural Network (ANN)

Levenberg-Marquardt learning algorithm

It resulted high performance 0.967 (the area under ROC curve) for the output from trained network related to real results.

An Artificial Neural Network Model for Neonatal Disease Diagnosis

Multi Layer Perceptron with a BP learning algorithm

Predictive accuracy acquired was 75%

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

Back-Propagation algorithm

Predictive accuracy obtained was 89%

18

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)

Methods:

41 clinical and laboratory parameters of 131 patients (63 of them were cancerous and others non-cancerous) who had pathological results were selected from patients’ documents from Sina hospital of Hamadan.

19

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)These are: age, gender, vomiting, nausea, hematocrit, W.B.C, ESR, infection, weight loss, M.C.H, M.C.H.C, Na+, MCV, RDW, PT, SGOT, SGBT, creatinine, uric acid, bilirubin D, bilirubin T, LDH, fever, hemorrhage, lymphadenopathy, hepato and splenomegaly, hemoglobin, platelet and so forth.

20

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)After primary statistical analysis, the two tailed student t-test was used to determine the statistical significance for the difference between the two groups of patients with and without Leukemia. Eight of 41 features which showed more significant difference between cancerous and noncancerous group were used as inputs for ANN analysis.

21

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)These features were: gender, fever, hemorrhage, lymphadenopathy, massive liver and spleen, hematocrit, hemoglobin, and platelet.

Assembling and training of ANN was done by Matlab software r2007b. In order to train neural network, selected features were normalized.

22

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)

After normalization a randomly chosen sample was divided into training (80%), cross validation (10%) and testing datasets (10%). The training data set was presented to the network for learning.

23

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)Multilayer perception model of the ANN was used. The network consists of an input layer, a hidden layer and an output layer. The input layer contained 8 neurons corresponding to eight input features; the hidden layer contained eight neurons transforming the input features from input layer to hidden layer. Finally, the output layer had only one neuron, representing two possible diagnosis states cancerous or noncancerous. Then the neural network was trained with the data on hand; learning function was LM (Levenberg- Marquardt back-propagation)

24

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)

Rsaults:

25

The ROC plot is merely the graph of points defined by sensitivity and (1 – specificity). Customarily, sensitivity takes the y axis and (1-specificity) takes the x axis. The sensitivity is simply the True Positive Fraction (TPF). In other words, sensitivity gives us the proportion of cases picked out by the test, relative to all cases that actually have the disease. Specificity is the ability of the test to pick out patients who do not have the disease. It is simply the True Negative Fraction.

26

area under roc curve for this analysis was 0.967.

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)

Phi coefficient value was 0.778 and its P value was 0.005.

27

Recognition and prediction of leukemia with Artificial NeuralNetwork (ANN)Conclusion

In order to improve the performance of training process we must use large sample size (more patients). Finally we used the weight and bias matrix value of trained network and assembled ANN structure to program the software for quick and accurate detection of cancer quantitatively.

28