Metaheuristic Assisted Deep Neural Network for Diabetic Mellitus Diagnosis via Proposed Texture Feature Descriptor

Diabetes mellitus (DM) indicates the group of metabolic disorders, which shares the hyperglycemia phenotype.The secondary pathophysiological changes are caused by the metabolic dysregulations related to DM in many organ systems that result in various difficulties liable for the mortality and morbidity linked with the disease.However, the early diagnosis of the disease is possible from the human tongue.Therefore, this work introduces a new DM diagnosis model that includes two major phases: (i) Feature extraction, and (ii) Classification.Initially, the input image is given to the feature extraction phase, in which the color feature, texture feature, and geometry features are extracted.The texture features include the Proposed Tri-region distance-based Local Binary Pattern (TRD-LBP) and Gray Level Co-occurrence Matrix (GLCM) features, whereas, the 13 geometry features are also included in this feature extraction phase.Subsequently, these extracted features are provided as the input to the Optimized Neural Network (NN), where the diagnosis takes place.To make the detection more precise, the training of NN is carried out by the Self Adaptive Particle Swarm Optimization (SAPSO) algorithm via tuning the optimal weights.Finally, the performance analysis is made among the adopted and the existing approaches for various measures like sensitivity, NPV, accuracy, FDR, specificity, FPR, precision, MCC, FNR, and F1-score, respectively.