OrthoXIC: A Fusion Approach for Multi-Implant Classification using X-ray Images

Orthopaedic surgical implants are vital for treating musculoskeletal disorders, necessitating precise implant identification in x-ray images to ensure accurate evaluation and management. Most of the studies focus on specific joint implants due to a lack of data availability and existing generic models suffer from the challenges of high similarity within and between different types of implants. To address these problems, a novel OrthoXIC method that automatically classifies orthopaedic multiple implants is proposed. To develop a generic multi-implant classification model, 546 x-rays with implants consisting of needles, screws, and plates of the elbow and wrist joints were collected. The canny edge detection and binary thresholding are applied to extract the specific implant region. The fusion mechanism extracts appropriate texture and deep features from modified HOG and VGG16, respectively. The Random Forest improves model efficiency by addressing the inter- and intra-class variation in the concatenated features and achieves a better result with 95% accuracy, 94% precision, 93% recall, and 93% f1-score for the collected dataset. The proposed model shows better generalization compared to state-of-the-art methods.