Knee osteoporosis detection from Deep learning stacking ensemble model on Preprocessed X rays Images

Abstract

Osteoporosis is a progressive skeletal disease characterized by a steady decline in bone mineral density (BMD) and deterioration of bone microarchitecture, ultimately resulting in fragile bones, higher fracture risk, and impaired quality of life. The current clinical gold standard for diagnosis, dual-energy X-ray absorptiometry (DXA), provides accurate BMD measurements but remains limited by high cost, restricted accessibility, and impracticality for population-level screening. These limitations have encouraged increasing interest in computer-aided diagnostic approaches based on plain radiographs, which are inexpensive, widely available, and non-invasive. In this study, we propose a deep learning framework for stage-wise classification of osteoporosis from knee radiographs, focusing on the clinically significant distinction between Normal, Osteopenia, and Osteoporosis. The framework integrates a robust preprocessing pipeline, transfer learning across multiple state-of-the-art convolutional neural networks, and a novel ensemble strategy that fuses predictions through weighted averaging and a static deep meta-learner. Experimental evaluations on publicly available datasets demonstrate that the ensemble outperforms individual models, achieving higher accuracy and balanced precision-recall scores across all classes. Beyond performance gains, the framework presents a scalable and clinically relevant solution to the limitations of existing diagnostic methods. These findings underscore the potential of ensemble deep learning to strengthen automated osteoporosis detection and enhance clinical decision-making.

Ähnliche Arbeiten