Hybrid Optimization Driven Deep Learning Classifier for Multilevel Spinal Cord Disease Classification

Abstract

Spinal cord is shape of cylindrical, which is positioned in Central Nervous System (CNS) and enlarged among lumbar vertebrae and medulla oblongata. The neural signals are passed through the brain and residual sections of body particularly by means of spinal cord, which acts as passageway. Moreover, spinal cord injury is a most distressing condition for humans and it may cause sensation loss and affects appropriate functions of musclespermanently or temporarily. In addition, the spinal cord injury blocked the nervous system, which restricts the body mobility and weaken the quality of life. Besides, the neurological circumstance of patient can be enhanced by early classification of disease and propertreatment of spinal cord disease. Therefore, the step to guarantee the recovery by early functioning and safeguard a suitable interceptive is essential. This research is developed with three contributions for multilevel spinal cord disease classification using medical images. In the first contribution, Crow Search-Rider Optimization-based Deep Convulution Neural Network (CS-ROA DCNN) for spinal cord segmentation and injury identification. Here, the thresholding approach is employed for segmentation process. Moreover, Sparse Fuzzy C-Means (Sparse-FCM) clustering technique is utilized to performed disc localization. The significant features are extracted, and it is passed to DCNN classifier for classification process. Meanwhile, the CS-ROA is developed in order to train the DCNN classifier for effective classification process. However, the developed CS-ROA is introduced by incorporating Crow Search Algorithm (CSA) and Rider Optimization Algorithm (ROA). In second contribution, Deep Recurrent Neural Network (DRNN) is devised for injury level categorization of spinal cord. In this method, adaptive thresholding process is employed for segmentation and Sparse FCM is utilized for disc localization. Furthermore, the features are extracted from a localized disc, whereas injury detection is carried out using CS-ROA-based DCNN tec