Fragility Fracture Risk Prediction through the integration of Artificial Intelligence and Radiofrequency Echographic Multi Spectrometry (REMS) Technology.

BACKGROUND AND

OBJECTIVE

Bone fragility is a major clinical issue and a leading cause of disability, especially in aging populations. Traditional diagnostics rely on bone mineral density, which does not fully capture structural and micro-architectural factors influencing fracture risk. This study investigates Deep Learning applied to ultrasound Radiofrequency Echographic Multi Spectrometry (REMS) data to improve fracture risk identification and stratification.

METHODS

A retrospective multicenter dataset of 3510 participants was analyzed. A Convolutional Neural Network based on AlexNet was trained in a supervised manner to classify subjects as non-fractured, at-risk, or fractured. Performance was evaluated via 5-fold cross-validation, with an ensemble of the best model from each fold to enhance robustness and generalization. A continuous Fragility Index (0-100) was derived from softmax probabilities to quantify risk and stratify patients into low-, medium-, and high-risk categories.

RESULTS

The model showed consistent performance at both frame and patient levels. For binary classification of at-risk versus not-at-risk individuals, the ensemble achieved approximately 80% sensitivity and 80% specificity, balancing detection and false positives. The Fragility Index offered a continuous, interpretable measure of fracture susceptibility, enabling effective risk stratification. Weak correlation with bone mineral density suggests that the model captures structural features beyond mineral content.

CONCLUSIONS

Deep Learning analysis of ultrasound Radio-Frequency signals provides a non-ionizing, portable approach for assessing bone fragility. This method may support early identification of at-risk individuals and enable personalized prevention and management strategies in osteoporosis care.