AI model improves thyroid cancer diagnosis

An artificial intelligence (AI)-based model, dubbed AI-Thyroid, for the diagnosis of thyroid cancer demonstrates its utility in improving diagnostic performance and interobserver agreement, which could be helpful in less-experienced physicians, reports a study.

A total of 19,711 images of 6,163 patients in a tertiary hospital (Ajou University Medical Center; AUMC) were used to train the new system. AI-Thyroid was validated using 11,185 images of 4,820 patients in 24 hospitals (test set 1) and 4,490 images of 2,367 patients in AUMC (test set 2).

To determine the clinical implications, the authors compared the findings of six physicians with different levels of experience (group 1: four trainees; group 2: two faculty radiologists) before and after AI-Thyroid assistance.

AI-Thyroid achieved an area under the receiver operating characteristic (AUROC) curve of 0.939. For test 1, its AUROC, sensitivity, and specificity were 0.922, 87.0 percent, and 81.5 percent, respectively. The corresponding values for test set 2 were 0.938, 89.9 percent, and 81.6 percent.

The AUROCs of AI-Thyroid did not differ significantly in relation to the malignancy prevalence (>15.0 percent vs ≤15.0 percent; p=0.226).

In the simulated set-up, AI-Thyroid assistance improved the AUROC, sensitivity, and specificity from 0.854 to 0.945, from 84.2 percent to 92.7 percent, and from 72.9 percent to 86.6 percent (p<0.001 for all) in group 1, and from 0.914 to 0.939 (p=0.022), from 78.6 percent to 85.5 percent (p=0.053), and from 91.9 percent to 92.5 percent (p=0.683) in group 2.

“The interobserver agreement improved from moderate to substantial in both groups,” the authors said.