Lightweight Edge-Aware Feature Extraction for Point-of-Care Health Monitoring

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Osteoporosis diagnosis from X-rays is difficult due to visual similarity between healthy and diseased bone structures.
• Current methods struggle to identify subtle indicators of osteoporosis, necessitating advanced image analysis techniques.

Purpose of the Study:

• To develop a novel framework for enhanced osteoporosis classification from X-ray images.
• To capture subtle structural differences not visible to the human eye using gradient-based maps.

Main Methods:

• Utilized analytic Gabor filters for multi-scale, multi-orientation image decomposition.
• Constructed filter response matrices and derived second-order texture features via covariance analysis and eigenvalue decomposition to create Gabor Eigen Maps.
• Processed feature maps with a convolutional neural network (CNN) for high-level descriptor extraction and subsequent classification using machine learning algorithms.

Main Results:

• The proposed Gabor Eigen Map framework significantly improved osteoporosis classification accuracy compared to existing methods.
• Demonstrated superior performance in identifying osteoporotic cases from X-ray images.
• The method effectively captures subtle textural and structural variations indicative of osteoporosis.

Conclusions:

• The novel framework offers a powerful tool for accurate and efficient osteoporosis detection from X-ray images.
• The Gabor Eigen Maps provide an interpretable and lightweight representation suitable for edge device deployment.
• The approach holds significant potential for real-time, privacy-preserving osteoporosis screening at the point of care.