Neuro-Immune Model Based on Bio-Inspired Methods for Medical Diagnosis

Introduction

The use of artificial intelligence techniques in healthcare informatics has gradually increased in recent years to solve various problems with the classification of medical diseases, which are a real challenge and a complex task for researchers. There is no doubt that the most important factors related to the disease are medical disease evaluation and expert decisions. Besides, there can be several problems with any type of real medical data, one being the problem of medical disease classification. Several studies have been conducted to achieve high performance using various research strategies for medical diagnosis. A study performed by (Tanwani & Farooq, 2009, 2010) shows that the classification accuracy depends on the complexity of the medical datasets chosen, not on the classifier model choice. This complexity of the medical disease dataset will be useful for researchers to evaluate the process of classification of their dataset for automatic knowledge extraction. Recently, bio-inspired approaches are probably one of the most active and popular research topics with broad multidisciplinary connections in the study of natural systems to find solutions to the classification problem of medical diseases that cannot be solved by classical methods (Xin-She & Karamanoglu, 2013). Indeed, the development of a new classification technique based on bio-inspired methods is essential to improve the medical disease diagnosis method. It is essential to hybridize different bio-inspired algorithms to provide innovative solutions to medical disease classification problems. A research study focused on the use of hybrid components for the classification of medical data is therefore needed (Al-Muhaideb & El Bachir, 2013).

This study aims to propose a hybrid approach based on bio-inspired algorithms called Neuro-immune, which combines multilayer perceptron neural network (MLPNN) and Clonal Selection Classification (CSC) for the medical data disease to provide higher classification accuracy also optimize their performance. The proposed algorithm is to convert the MLPNN concatenation vector weights from the training samples into an antigenic vector, and then decompose it into new weights to generate a population of memory cells. Then it is used by the cloning and mutation processes of the CSC algorithm to select the best candidates of antibodies so that one of them will contain the next MLPNN weights. After several iterations, update the MLPNN weights by minimizing the mean square error using the back-propagation algorithm and the last update of the weights represents the optimal value of the final structure of the network that will be used for the classification of disease types with a high classification rate. The approach proposed in this paper has been evaluated on four sets of benchmark medical data (Breast Cancer, Hepatitis, Diabetes, and Parkinson) collected from the UCI machine learning repository (Dua & Graff, 2019), which are the riskiest and frequent diseases of the century. The Classification accuracy, specificity rate, and sensitivity rate have been used as well as 4-fold cross-validation to assess the reliability of the proposed approach.

This work makes it possible to use the advantages of the artificial immune system which are the capabilities of recognition of new types and the memorization that can be trained as an excellent tool for learning and classification optimizing the performance of the multilayer neural network Perceptron for a correct and effective diagnosis of different diseases. Furthermore, this property is the only advantage of the CSC algorithm over the MLPNN.

The remainder of the paper is organized as follows: Section 2 presents some related work on medical disease classification. Section 3 provides an overview of the algorithms used. Section 4 describes in detail the neuro-immune algorithm proposed. Section 5 presents the experimental results and the paper concludes in Section 6.