A deep-learning model trained on chest x-rays significantly outperformed the Fracture Risk Assessment Tool (FRAX) in predicting major osteoporotic fractures, according to this presentation.
Led by presenter Yisak Kim, a doctoral student at Seoul National University in South Korea, the developers say the AI approach could provide a more accessible and accurate tool for fracture risk assessment in clinical practice.
FRAX is widely used for predicting fracture risk but is limited by its dependence on dual-energy x-ray absorptiometry (DEXA) exams and low responsiveness to short-term fracture events, according to the researchers. Hence, they aimed to develop and validate a deep-learning (DL) model based on chest x-rays to predict major osteoporotic fractures and to evaluate its predictive performance relative to FRAX.
The group used chest x-rays from 42,014 patients (mean age, 59.3 years; 79.8% women) from one institution to develop the model and 31,821 from patients (mean age, 64.6 years; 72.4% women) from a second institution for external validation. The model’s predictive performance was evaluated using Concordance index (C-index) and time-dependent area under the receiver operating characteristic curve (AUROC) at two, three, five, and 10 years.
In the external test set, the model yielded a C-index of 0.78 compared with 0.767 for FRAX (p < 0.001). The two-, three-, five-, and 10-year AUROCs for the DL model were 0.804, 0.816, 0.831, and 0.865 compared to AUROCs 0.774, 0.78, 0.784, and 0.810 for FRAX (all, p < 0.001). The DL model also demonstrated strong performance across fracture subtypes: C-indices of 0.828 (vertebral), 0.71 (nonvertebral), and 0.833 (hip).
Ultimately, the DL-based approach enables fracture risk prediction from widely available chest x-rays, and this offers a practical and accessible alternative to DEXA-based assessments, according to the group.
Fill out your Tuesday afternoon and find out all the details in this session.