Calibration and Prospective Validation Needed for <scp>AI</scp> ‐Assisted Thyroid Nodule Diagnosis

Abstract

To the Editor: We read with great interest the article by Abbasian Ardakani et al "Diagnosis of thyroid nodule malignancy using peritumoral region and artificial intelligence," 1 in which the authors develop and externally validate a hybrid radiomics model that incorporates peritumoral features to distinguish benign from malignant thyroid nodules on ultrasound.The study's robust design and highly reported area under the curve values are commendable, yet we wish to highlight two aspects that could further strengthen its translational impact.First, the authors focus on discrimination but do not report model calibration or assess clinical utility.Without calibration metrics-such as calibration plots or the Hosmer-Lemeshow test-and decision-curve analysis to evaluate net benefit across threshold probabilities, it remains unclear whether the predicted malignancy probabilities align with observed outcomes or will improve patient management in practice. 2,3Future work should include these evaluations to ensure that the model's risk estimates are both accurate and actionable, and to identify the threshold at which application of the AI tool would yield maximum clinical benefit.Second, although external validation across multiple centers is a strength, prospective assessment within routine clinical workflows is lacking.Retrospective image selection and offline analysis may overestimate performance when compared to real-time deployment by sonographers, and integration into reporting systems could introduce operational challenges. 4A prospective implementation study, assessing the tool's impact on biopsy rates, diagnostic yield, and reading times, would be critical prior to widespread adoption.We congratulate the authors on their rigorous multicenter study and encourage future efforts to incorporate calibration, decision-curve analysis, and prospective workflow integration to fully realize the promise of AI-assisted thyroid nodule evaluation.

Ähnliche Arbeiten