This study combines the Radiofrequency Echographic Multi-Spectrometry (REMS) approach with deep learning to analyze bone micro-architecture and predict fragility fracture risk using hip ultrasound (US) scans. The dataset includes 2,084 US scans, comprising 352 from subjects with prior fragility fractures and 1,732 from non-fractured individuals at the time of scanning. To evaluate the model’s predictive capability, fracture status in the test set was determined five years after data acquisition. The algorithm, trained on numerical matrices extracted from US images, detects micro-structural changes linked to fractures, independent of bone mineral density (BMD). The preliminary results show that the CNN-based model classifies subjects with up to 80% accuracy based on the fracture history. Expanding the dataset, particularly by increasing the number of fractured cases, could further improve model sensitivity. These findings underscore the potential of deep learning in US imaging for bone fragility risk assessment, offering an alternative to BMD-dependent methods.
Estimation of Bone Fracture Risk by combining Artificial Intelligence techniques and Radiofrequency Echographic Multi Spectrometry
Peluso G.;Epicoco I.;Cafaro M.;
2025-01-01
Abstract
This study combines the Radiofrequency Echographic Multi-Spectrometry (REMS) approach with deep learning to analyze bone micro-architecture and predict fragility fracture risk using hip ultrasound (US) scans. The dataset includes 2,084 US scans, comprising 352 from subjects with prior fragility fractures and 1,732 from non-fractured individuals at the time of scanning. To evaluate the model’s predictive capability, fracture status in the test set was determined five years after data acquisition. The algorithm, trained on numerical matrices extracted from US images, detects micro-structural changes linked to fractures, independent of bone mineral density (BMD). The preliminary results show that the CNN-based model classifies subjects with up to 80% accuracy based on the fracture history. Expanding the dataset, particularly by increasing the number of fractured cases, could further improve model sensitivity. These findings underscore the potential of deep learning in US imaging for bone fragility risk assessment, offering an alternative to BMD-dependent methods.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
