Sparse image reconstruction for distal radius fracture

 There are several treatment options for distal radius fractures, with the current primary method using volar locking plates (VLP), which account for 83% of all surgically treated distal radius fractures requiring internal fixation. However, specific fracture patterns, such as very distal volar rim fragments, volar ulnar corner fragments, and dorsal ulnar corner fragments, require surgeons to choose implants other than VLP. Preoperative computed tomography (CT) scans can effectively identify whether intra-articular distal radius fractures extend into the sigmoid notch, a critical factor influencing the surgeon's decision to use fixation methods other than VLP. Therefore, accurately assessing the size of a volar ulnar corner or dorsal ulnar corner fragments preoperatively through CT imaging is crucial for selecting the appropriate surgical implants and technique. 

However, arranging CT scans is often costly, time-consuming, and may not provide timely images for the surgeon. In many cases, surgeons only need a few cuts of the joint surface fracture, making a full CT scan underutilized. If surgeons could assess the condition of the intra-articular fracture without exactly performing a CT scan, it would save significant healthcare costs, time, and radiation exposure.  

Using a diffusion model, our research team developed an AI-based solution to reconstruct tomographic and 3D images from three standard X-ray views (AP, lateral, and oblique). Initial clinical validation, conducted by three hand surgeons from Massachusetts General Hospital in Boston and Kaohsiung Medical University Hospital in Taiwan, has shown that the reconstructed images have minimal artifacts compared to the original CT scans. These reconstructed images can assist clinicians in better understanding the structure of intra-articular fractures. The model will benefit from further refinement through collaboration with more medical centers worldwide, utilizing federated learning to improve its accuracy and robustness over time.