Fusion of ViT-Based Deep Features and Phi-Grid Handcrafted Features for Osteoporosis Detection Using Attention-Infused ResNet MLP and Autoencoder

Abstract

Osteoporosis is a progressive skeletal disorder characterized by reduced bone mass and deterioration of bone microarchitecture, significantly increasing fracture risk, particularly among the elderly. Early detection of osteoporosis is vital for preventing fractures, reducing healthcare burdens, and improving patient quality of life. Medical imaging, particularly X-ray analysis, offers a non-invasive and accessible approach for screening. In this research, we propose a robust feature extraction and classification framework for osteoporosis detection using publicly available knee X-ray osteoporosis datasets. We employ a Phi-grid based block division method to partition X-ray images into meaningful regions, from which handcrafted local features (texture, frequency, statistical, spatial, and structural descriptors) are extracted. These handcrafted features are combined with Vision Transformer (ViT)-enabled features to effectively capture both local and global image representations. The resulting feature sets, considered individually and in combination, are fed into a deep learning-based hybrid GAMLP-Net classifier, which integrates ResNet MLP and Autoencoder blocks enhanced with attention mechanisms to improve feature representation and classification robustness. The model performs classification for both binary class (Normal vs. Osteoporosis) and multiclass (Normal, Osteopenia, Osteoporosis) tasks. Experimental results demonstrate that the combined feature representation significantly improves classification accuracy and outperforms several state-of-the-art methods, validating the effectiveness of the proposed approach for automated osteoporosis diagnosis.

Ähnliche Arbeiten