Fracture Healing Assistant

 

AI-assisted quantitative analysis of fracture healing and prognosis of musculoskeletal healing processes

Most bone fractures heal without complications. However, in around 10–12 per cent of cases, the bone fails to heal properly, resulting in pseudarthrosis. This can cause pain, reduced mobility and, in many cases, an inability to work for months or even years. Until now, doctors have assessed fracture healing using serial X-rays—a subjective process that depends heavily on clinical experience. As a result, critical cases are not always identified early enough.

The collaborative project Fracture Healing Assistant is developing AI-based software that will enable doctors to assess and predict fracture healing objectively and quantitatively. The software automatically detects the fracture site, tracks healing progress spatially and over time across multiple images, and uses these data to calculate a healing score and an individual risk prognosis. It is being designed to integrate seamlessly with existing X-ray equipment and picture archiving and communication systems (PACS), as well as with everyday clinical workflows. The project aims to identify healing complications an estimated six weeks to three months earlier than conventional methods, allowing treatment decisions to be made sooner and on the basis of objective evidence. It brings together partners from Schleswig-Holstein, Hamburg and Mecklenburg-Vorpommern in a cross-state collaboration across northern Germany, combining expertise in medicine, engineering, computer science and industrial software development.

At the IME, the AI algorithm for determining risk scores is being developed as a leading effort. The model processes serial X-ray images alongside further features derived from the upstream image analysis AI, and integrates additional clinical parameters such as age, weight and gender. 

A key area of research at the IME is the reliability and transparency of these models. Building on the foundations laid by previous projects, the team is investigating methods for quantifying uncertainty and ensuring explainability. The aim is to enable the system to communicate the basis for its decisions transparently and provide reliable feedback in ambiguous cases. This is also essential for meeting the regulatory requirements of the Medical Device Regulation (MDR), the General Data Protection Regulation (GDPR) and the EU AI Act. In addition, the IME is contributing its expertise to the design of the image-analysis network architecture, the modelling of doctor–AI interaction and the clinical evaluation of the system.

Short Facts