Comparison of an AI-assisted system (S-detect) and sonographers of different experience levels in diagnosing thyroid...

Thyroid cancer (TC), the most common neck malignancy, can metastasize early. Conventional ultrasound diagnosis relies on subjective feature interpretation. Objective tools are needed to improve diagnostic efficiency. Objective: To compare the diagnostic efficacy of artificial intelligence-assisted ultrasonography (S-Detect) versus sonographers of varying experience in differentiating benign from malignant thyroid nodules.

This retrospective study analyzed 315 thyroid nodules (237 patients) undergoing ultrasound and biopsy/surgical confirmation. Sonographers were classified as junior or advanced. The diagnostic performance (sensitivity, specificity, accuracy, kappa, Youden's Index, AUC) of S-Detect and both sonographer groups was compared.

In the junior group (115 nodules), S-Detect outperformed junior sonographers (sensitivity 98.4% vs 96.9%, specificity 78.4%vs 52.9%, accuracy 89.6% vs 77.4%, kappa 0.784 vs 0.521, AUC 0.884 vs 0.749; all P<0.05) In the advanced group (200 nodules), S-Detect sensitivity (97.5%) matched senior sonographers (96.7%), but with lower diagnosis specificity (57.7% vs 69.2%). Senior sonographers showed higher accuracy (86.0% vs 82.0%) and kappa (0.691 vs 0.593), Compared with senior physicians, S-Detect demonstrated comparable diagnostic efficacy to the senior group in identifying malignant nodules, while showing slightly inferior performance to senior ultrasound specialists in diagnosing benign nodules. Senior physicians exhibited superior accuracy and consistency in nodule diagnosis compared to S-Detect; however, no significant difference was observed between the two in overall performance (P > 0.05).

S-Detect surpasses junior sonographers in diagnosing thyroid nodules. Its overall diagnostic performance is comparable to advanced sonographers.