A novel deep learning model significantly enhances the preoperative diagnosis of follicular-patterned thyroid neoplasms (FNs) using routine ultrasound images. In extensive multicenter validation, the model achieved an impressive AUC of 0.937 for internal and 0.853 for external cohorts. Moreover, it demonstrated high accuracy (90.9% and 82.8%) and sensitivity (93.9% and 84.5%), streamlining clinical decision-making and minimizing unnecessary treatments. This development marks a pivotal shift towards more precise, non-invasive thyroid diagnostics.
Journal Article by Shen H, Pei S (…) Zhang B et 23 al. in EClinicalMedicine
© 2025 The Authors.