Authors: Auti Om Jitendra, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Dr. A. A. Khatri
Abstract: Bone fractures are among the most common musculoskeletal injuries, and their outcome depends on accurate classification, timely localisation and early recognition of failed healing. Manual radiograph interpretation is subjective and time-consuming, and plain X-ray is slow to reveal healing failure (nonunion), which affects up to 12 % of tibial fractures and can take more than six months to diagnose [1]. Artificial intelligence (AI), combined with medical imaging and novel sensing, is increasingly proposed as a route to fast, consistent and early fracture assessment. This paper presents a focused narrative review of four verified peer-reviewed IEEE Access studies (2022–2025) covering contactless antenna-based sensing of implant deflection and fracture stiffness, X-ray fracture-type classification with transfer learning and ensemble classifiers, two-stage CT-based cervical spine fracture classification and vertebra detection, and attention-enhanced YOLOv8 detection of paediatric wrist fractures. The review examines three connected themes: imaging and sensing technologies (X-ray, computed tomography and near-field electromagnetic sensing), learning methods (CNN transfer learning, hybrid CNN–classical ensembles and object detectors), and clinical integration with explainability. The literature shows that deep models can classify fracture types with about 95–98 % accuracy, that YOLOv8 variants reach about 93.5 % mAP50 for vertebra detection and 65.8 % mAP50 over all wrist-trauma categories, and that a miniaturised Vivaldi antenna (36 × 30 mm) can track sub-millimetre implant displacement. However, most studies use single-source datasets, small test sets, random data splits and offline evaluation, and few report clinical validation. The review contributes a comparative analysis of representative studies, a proposed conceptual architecture linking acquisition, AI inference, decision support and clinical reference validation, and a prioritised agenda of research gaps and future directions.
