gms | German Medical Science

Objectives: Early detection of pseudarthrosis is crucial to be able to initiate countermeasures in due time. Existing scoring systems for risk of non-unions lack precision due to subjective interpretation, variability among surgeons [1] and non-consideration of load bearing patterns, potentially leading to misdiagnosis or delayed treatment.

We propose a novel pseudarthrosis prediction model that combines mechanobiological computer simulations (digital twin) with machine learning to improve predictive accuracy compared to conventional scoring systems based solely on clinical and surgical parameters.

Methods: A dataset of 100 tibia shaft fracture cases was analyzed. Case data included post-OP bi-planar X-rays as well as parameters such as patient side, age, weight, sex, degree of soft tissue damage, fracture type, allergies, comorbidities and the used implant type and size. The clinical endpoint was defined as either consolidated or non-union.

From the bi-planar post-OP radiographs we reconstructed 3D volume models of the patient-specific anatomy and fracture geometry. As aftercare is considered in the digital twin, we assumed full weight bearing to predict individual healing using an established mechanobiological fracture healing model [2]. For reference purposes we evaluated all cases using the LEG NUI scoring system [3].

Results and conclusion: Applying the LEG-NUI scoring system to our dataset correctly identified only a single non-union case (Figure 1 [Fig. 1]), resulting in correspondingly low sensitivity as well as low precision (PPV) (Table 1 [Tab. 1]).

In contrast to that the proposed model that combines the output of the mechanobiological simulations with clinical factors correctly identified 16 out of the 21 pseudarthrosis cases (Figure 1 [Fig. 1]). Compared to the LEG NUI-based prediction, sensitivity increases from 5% to 76% and precision (PPV) from 20% to 62% (Table 1 [Tab. 1]).

In this study we demonstrated that augmenting scoring systems with digital twin-based predictive analytics has the potential to significantly improve the early prognosis of pseudoarthosis. In addition, digital twins have the potential to add a scenario-based analysis of the weight bearing capacity during aftercare. Further research should evaluate the predictive performance of other available metrics (e.g. NURD, TFHS) as well as experience-based surgeon ratings.