Accurate Osteoporosis Diagnosis Model Proposing Genetic Algorithm Optimization for Convolutional Neural Network

Abstract

Osteoporosis is characterized by diminished bone mass and bone tissue loss, which results in weakened bones, decreased bone strength, and increased risk of fractures. This paper exploits a Medical Lumber Spine Images (MLSI) of Dual-Energy X-ray Absorptiometry (DEXA) clinic in Mosul / Iraq to be classified as either normal or having osteoporosis. It also presents the capability of optimizing the hyperparameters of a deep learning model, where the Genetic Algorithm (GA) is used for optimizing Convolutional Neural Network (CNN) hyperparameters. The proposed model essentially explores and optimizes 18 hyperparameters; it is named the Genetic Optimization for CNN (GOCNN). It combines the powerful of GA and CNN in order to provide best hyperparameters tuning, this would further decrease the manual tuning efforts. The real clinical dataset of MLSI is utilized and employed in this study. The proposed method can correctly diagnose 93% of osteoporosis cases from unseen data, with an Area Under Curve (AUC) of Receiver Operating Characteristic (ROC) equals to 0.98.