Back to 2025 Abstracts
Optimizing Hand Splint Fitting Parameters through Artificial Intelligence Analysis
Ashkan Sedigh, MS
1; Peter Beredjiklian, -
1; Amir R Kachooei, MD
2; Michael Rivlin, MD
31Thomas Jefferson University, Philadelphia, PA; 2Rothman Orthopaedics Florida at AdventHealth, Orlando, FL; 3Rothman Orthopaedic Institute, Philadelphia, PA
Introduction: The current method for determining proper splint/brace fit involves measuring wrist circumference to classify as small, medium, or large, etc. This study evaluates hand features using 3D scanned data and Artificial Intelligence (AI) to improve the fit of pre-fabricated wrist splints. We hypothesize that AI generated sizing provides greater accuracy and enhance patient fitting outcomes. By incorporating advanced 3D scanning technology, we aim to develop a more personalized and effective approach to splint fitting.
Methods: We recruited healthy volunteers to participate in the study, resulting in a total of 54 hands being scanned. Each volunteer was fitted with wrist braces, and 3D data of their hands were collected using an infrared-based 3D scanner. The 3D scanned data were then analyzed to identify and measure 15 distinct hand and forearm features (Fig. 1A). We have used the correlation coefficients (Fig. 1B) between measured hand features and splints size to understand which features were most strongly correlated with splint size categories (small, medium, large). Subsequently, we developed a classification algorithm to predict the appropriate splint size based on the correlated hand features. We utilized three different machine learning models for this purpose: XGBClassifier, RandomForestClassifier, and Support Vector Classifier (SVC). Each of these classifiers was trained and evaluated to determine their accuracy and effectiveness in predicting the correct splint size.
Results: The results indicate that the hand wrist width (C) achieved the highest classification accuracy of 91% for both the XGBClassifier and RandomForestClassifier. The feature set including hand wrist width (C), mid-forearm width (E), and hand crease line width (A1) also performed well with the XGBClassifier, achieving the same accuracy of 90%. The SVC classifier showed consistent performance across various feature sets, with the highest accuracy of 81% for the feature sets. Overall, these findings suggest that hand wrist width (C) is the most predictive feature for splint size classification, with additional features (E and A1) providing minimal enhancement (Fig. 1C).
Conclusion: Using artificial intelligence and 3D scanning, hand-wrist width (C) was the most reliable predictor for determining splint size, achieving the highest accuracy across multiple models. Additional features such as mid-forearm width (E) and hand crease line width (A1) provide minimal improvement to predictive performance. Artificial intelligence can assist orthotists and physicians in optimal brace sizing with high accuracy from a single image, enabling contactless fitting for patients.
Back to 2025 Abstracts
