Combined CT radiomics and nnU-Net model improves calcaneal spur prediction

A CT-based model combining radiomics, nnU-Net deep features, and Gissane's angle predicted calcaneal spurs with AUC 0.772 (95% CI 0.700-0.843) in a test set of 161 CTs, outperforming clinical, radiomics, and nnU-Net models alone.

• The combined model incorporated a radiomics score, an nnU-Net deep feature score, and Gissane’s angle; age was excluded.
• In the test cohort, AUC improved from 0.691 (clinical) to 0.710 (radiomics) to 0.705 (nnU-Net) to 0.772 (combined).
• The retrospective study randomly split 803 calcaneal CT images into training (n=642) and test (n=161) cohorts for internal validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.