The following is a summary of “Artificial Intelligence Model Assisting Thyroid Nodule Diagnosis and Management: A Multicenter Diagnostic Study,” published in the February 2024 issue of Endocrinology by Ha, et al.

For a study, researchers sought to develop and validate an artificial intelligence (AI)-based model, AI-Thyroid, to diagnose thyroid cancer and assess its impact on diagnostic performance.

The AI-Thyroid model was trained using 19,711 images from 6,163 patients at a tertiary hospital (Ajou University Medical Center; AUMC). Validation was conducted using 11,185 images from 4,820 patients in 24 hospitals (test set 1) and 4,490 images from 2,367 patients at AUMC (test set 2). Clinical implications were evaluated by comparing the diagnostic findings of six physicians with varying experience levels (group 1: 4 trainees, group 2: 2 faculty radiologists) before and after AI-Thyroid assistance.

AI-Thyroid achieved an area under the receiver operating characteristic (AUROC) curve of 0.939. For the test set 1, AUROC, sensitivity, and specificity were 0.922, 87.0%, and 81.5%, respectively, and for test set 2, AUROC, sensitivity, and specificity were 0.938, 89.9%, and 81.6%, respectively. AI-Thyroid’s AUROC did not significantly differ based on malignancy prevalence (>15.0% vs ≤15.0%, P = .226). In simulated scenarios, AI-Thyroid assistance significantly improved AUROC, sensitivity, and specificity from 0.854 to 0.945, 84.2% to 92.7%, and 72.9% to 86.6% (all P < .001) in group 1, and from 0.914 to 0.939 (P = .022), 78.6% to 85.5% (P = .053), and 91.9% to 92.5% (P = .683) in group 2. Interobserver agreement improved from moderate to substantial in both groups.

AI-Thyroid enhanced diagnostic performance and interobserver agreement in thyroid cancer diagnosis, particularly benefiting less-experienced physicians.

Reference: academic.oup.com/jcem/article-abstract/109/2/527/7250484