Hybrid Framework between Bobcat Optimization and CNN for Osteoporosis Detection

Abstract Osteoporosis is characterized by diminished bone mass and bone tissue loss, which results in weakening bones, decreasing bone strength and increasing risk of fractures. This paper exploits a Medical Lumber Spine Images (MLSI) dataset of Dual-Energy X-ray Absorptiometry (DEXA) in a clinic in Mosul / Iraq in order to be classified as either normal or having osteoporosis. Here, it is decided to explore use of Bobcat Optimization Algorithm (BOA) for selection of best Convolutional Neural Network (CNN) hyperparameters. The proposed method is called Bobcat Optimization for CNN (BOCNN) model, which systematically explores and optimizes 18 hyper-parameters. It combines the effectiveness of BOA for optimizing effective CNN hyperparameters which can further reduce the manual tuning efforts. This study uses the real clinical data set from the MLSI. The proposed model is able to diagnose Osteoporosis correctly in 95% of the cases from unseen data. The Area Under Curve (AUC) of Receiver Operating Characteristic (ROC) equals to 0.98 Keywords: Osteoporosis, Deep learning, Bobcat Optimization Algorithm, Convolutional Neural Network