The role of artificial intelligence in surgeon-performed ultrasonographic evaluation of cytologically indeterminate thyroid nodules.

INTRODUCTION

Evaluating indeterminate thyroid nodules(ITN) is challenging, especially without molecular tests. This study examines whether artificial intelligence (AI) assistance can improve ITN diagnostic accuracy and bridge expertise gaps in surgeon-performed ultrasound.

METHODS

134 ultrasound clips from 67 patients with ITN were reviewed by doctors of four levels: endocrine-surgery specialist, senior residents, junior residents, and medical student. After a 2-week wash-out, they re-evaluated the clips using AI-SONIC, an AI platform analyzing ultrasound real-time to predict cancer risk. Performance was validated against final histopathology.

RESULTS

Without AI, medical students, junior residents and senior residents performed significantly worse than specialists(AUROC 0.530-0.560 vs 0.771, p ​< ​0.05). AI-SONIC improved residents' and medical students' diagnostic accuracy to levels comparable with specialists(AUROC 0.733-0.751 vs 0.771). The specialists' performance remained unchanged with AI assistance.

CONCLUSION

AI enhances ultrasound evaluation of ITN by junior surgeons and medical students, elevating their accuracy to expert levels, supporting clinical assessment and medical education.