Congratulations to MDRI members (PhD Candidate Daniela Mini and her Supervisors Prof Karen Reynolds and Prof Mark Taylor) who recently published a paper to share groundbreaking findings from their latest study on proximal humerus fracture fixation! 🦴🔩🔍

Focused on the impact of screw length on bone strain using advanced adaptive neural network (ANN)-based surrogate models, their innovative approach allows them to predict bone strain with high accuracy while significantly reducing computational costs.

These findings can revolutionise surgical practices by optimising screw configurations, potentially reducing the risk of fixation failure and improving patient outcomes.

🔗 Read More: https://pubmed.ncbi.nlm.nih.gov/38866503/

Simplified Abstract

Fracture plates used in shoulder bone fractures often fail. We don’t fully understand why because it’s complicated. Factors like the number, position, length, and angle of screws make it hard to pinpoint the exact reasons.

We use a method called finite element (FE) analysis to study these plates, but it’s expensive and can’t test every possible screw combination. Instead, we can use a simpler method called surrogate modeling, which is cheaper and faster.

In this study, we developed a model using adaptive neural networks (ANN) to predict how changes in screw length affect bone strain. We trained the ANN models with data from FE simulations conducted on a single bone, varying the length of the screws. After finding the best amount of training data, we trained different models.

The best ANN model was able to predict all possible screw lengths and was very accurate when compared to new FE data, with a high correlation (R2 = 0.99) and low error rates (0.51%-1.83% strain). It showed that the screw providing medial support had the most impact on the strain.

Overall, the ANN-based model was accurate and could be used for more complex cases with more variables.