After almost 25 years of basic research in biomechanics, in 2022 we built a proof of concept of our idea of a clinical bone fracture management software with our clinical colleagues from the University Hospital Ulm. In our recent project together with a group of BG Clinics, we moved into tibial fractures and a substantially larger data set with a lot of patient data to play with. The results are very promising and we want to present some of the insights for discussion. Beware, we won’t spoil the numbers game (publication pending)!
Building on the established Ulm bone healing model, our simulation predicts the healing progress over time. With the study design, we addressed some of the limitations of the approach:
-
Mechano-biological simulation covers reposition and retention of surgical treatment of bone fractures, but hardly accounts for any human biological factors.
-
Vascularization, which is crucial for the supply of nutrients to the fracture, is only considered in a very simplified form.
-
Co-morbidities, patient compliance or other patient parameters are not part of the model.
For example, a perfectly repositioned fractured bone, a perfectly executed osteosynthesis and an healthypatient without known comorbidities in the simulation should results in a healing bone. So far, factors other than biomechanical ones are only used to explain deviations in simulation results. We are about to change that.
To capture the complexity of patient data, including non-mechanical influences, we have therefore built a first AI-assisted simulation model. Detecting potential pseudarthrosis early in the treatment process could help clinicians implement an adapted treatment process and, most important, save patients a lot of pain.
Figure 2: Performance comparison of the two prediction approaches for detecting patients at risk for developing pseudarthrosis. Red graph shows the performance of the purely mechano-biological simulation, while the blue graph depicts the new AI-assisted simulation approach.
The red graph indicates that the purely mechano-biological simulation already generates meaningful predictions and is a reasonable base for further development. A variation of the method already creates value for surgeons within the OSORA education tool, where the concept of stability and its influence on bone healing can be trained.
Predictions based on our novel AI-assisted simulation also consider the wealth of auxiliary patient data. The blue graph thus indicates better improved sensitivity (= true positive rate) and specificity (false positive rate = 1 – specificity). The new approach shows highly motivating new aspects:
-
The method results in fewer false-positives, meaning less healing patients are classified as non-healers by the prediction. Translated to clinical reality, unnecessary additional examinations or follow-ups for the patient can be avoided.
-
Less false-negatives, meaning less non-healers are declared as healers. Instead, the technology helps identifying patients that are actually at risk for non-union, which could then be tracked closer during the rehabilitation phase.
The results are very promising for bringing the technology into clinical practice. As promised in the headline, here are our five reasons why we believe that AI-assisted simulation is the way to go:
-
Adding data driven improves the prediction results of the OSORA bone healing simulation.
-
Patient-specific prediction of the healing process enables for truly tailor-made treatment approaches if necessary.
-
Clinicians have more information for their decision making in therapy planning.
-
The OSORA technology enables risk-free testing of treatment approaches on a patient’s digital twin.
-
From a technical perspective, pushing usability for future clinical application of the product is the next step. Regulatory pathway and simulation performance are two of our next challenges.
We are working hard on bringing the technology into clinical practice. Thanks to all the clinical project partners involved: Your user feedback pushes us every day!