Deep Feature Fusion of CNN and Vision Transformer for Automated Osteoporosis Detection from Knee X-Ray Images

The overall contribution of this research work is as follows:
• Identification of gaps in existing method for knee osteoporosis detection, including limited dataset size, low
images quality, and insufficient feature extraction in conventional (CNN) approaches.
• Application of sophisticated data augmentation methods
to enhance model robustness and generalization on constrained datasets.
• Develop an ensemble based deep learning model that
can do automated diagnostic system using DenseNet169
with a Vision Transformer–inspired attention mechanism,
enabling the model to focus on clinically relevant regions
of knee X-rays.