Hybrid Optimization Model Integrating Gradient Descent and Stochastic Descent for Enhanced Osteoporosis and Osteopenia Recognition

Abstract

Osteoporosis and osteopenia, prevalent bone diseases affecting millions of people globally, necessitate accurate early diagnosis for effective treatment and fracture prevention. This paper proposes a novel hybrid optimization algorithm tailored for classifying these conditions based on Bone Mineral Density (BMD) measurements. The algorithm, a customized Mini-Batch Gradient Descent (MBGD), blends the advantages of Gradient Descent (GD) and Stochastic Gradient Descent (SGD), addressing specific needs for osteoporosis and osteopenia classification. Utilizing a dataset comprising BMD measurements and clinical risk factors from the Osteoporotic Fractures in Men (MrOS), Study of Osteoporotic Fractures (SOF), and Fracture Risk Assessment (FRAX), the model achieves an impressive accuracy of 99.01%. The proposed model outperforms existing methods, demonstrating superior accuracy compared to the accuracy obtained in Gradient Descent of 97.26%, Stochastic Gradient Descent of 97.23%, and other optimization algorithms such as Adam of 96.45% and the RMSprop of 96.23%. This hybrid model presents a robust framework for early diagnosis of Osteoporosis and osteopenia, and hence there is an enhancement in quality of life.