Weighted Feature Extraction for Bone Fracture Detection in X-Ray Images Using Modified Glcm and Machine Learning Methods

Resumen

Detecting bone fractures are an important use of medical imaging, which needs reliable extraction of features for the classification and diagnosis of fractures. The Gray Level Co-occurrence Matrix (GLCM) is a common tool used to achieve texture analysis by computing relationships among pixel intensities in space. Nevertheless, these methods tend to lose sensitivity to subtle but important changes in intricate structures, such as fractures. In this work, we provide a coded GLCM methodology for feature extraction in fracture detection. Improvements include: adaptive weightings based on the distance of pixels; normalization based on local contrast differences; and incorporation of additional statistical features. These enhancements also provide a statistical basis for emphasizing texture irregularities relevant to fractures, and down-weighting noise and redundancies in many extracted features. Comparisons of our modified GLCM with traditional GLCM demonstrate enhanced texture characterizations and classification accuracy across fracture detection in medical images. Our modification integrates texture analysis, which includes GLCM, with domain-specific needs for the medical imaging community, and offers an alternative, reliable approach for automated classification and diagnosis of bone fractures.