Diagnostic value of a dynamic artificial intelligence-based, ultrasound-assisted diagnostic system in differentiating between benign and malignant thyroid nodules with a diameter greater than 2 cm.

Background: In the diagnosis of larger thyroid nodules, the internal structure of the nodules and the surrounding tissues are more complex, making the nodule characteristics atypical. The tissue structure may interfere with the ultrasound examination, increasing the risk of misdiagnosis or missed diagnosis. Meanwhile, the uneven distribution of malignant tissues in large nodules or the tendency of large nodules to undergo necrosis and liquefaction due to their size can lead to a higher false-negative rate in fine needle aspiration cytology (FNAC). This study aimed to evaluate the clinical utility of a dynamic artificial intelligence (AI)-based, ultrasound-assisted diagnostic system in differentiating between benign and malignant thyroid nodules with a diameter >2 cm, and its guiding significance for subsequent management. Methods: The surgical data of 276 patients with 297 thyroid nodules who underwent surgery were collected. Dynamic AI was used to distinguish between the benign and malignant thyroid nodules. The dynamic AI, preoperative ultrasound, preoperative FNAC, computed tomography (CT), and postoperative pathological results of the nodules were compared to evaluate the diagnostic efficacy of dynamic AI. Results: Dynamic AI achieved 86.2% accuracy, with a sensitivity of 86.2% and a specificity of 86.2%, and showed high consistency with the postoperative pathological results (kappa =0.723, P<0.001). Dynamic AI also had good stability. Dynamic AI had higher consistency with the postoperative pathological results than the Thyroid Imaging Reporting and Data System (TI-RADS) of the American College of Radiology (ACR), which relies on traditional ultrasound before surgery (kappa =0.567, P<0.001), as well as significantly higher specificity and accuracy, a higher positive predictive value (PPV), and a lower misdiagnosis rate (P<0.05). There were no significant differences between dynamic AI and FNAC in terms of sensitivity, accuracy, the negative predictive value (NPV), and the missed diagnosis rate. Compared with CT, dynamic AI had higher diagnostic efficacy, and significantly higher sensitivity and specificity, as well as significantly higher accuracy, a significantly higher PPV and NPV, and a lower misdiagnosis rate and missed diagnosis rate (P<0.05). There was no significant difference in diagnostic accuracy among the three non-invasive examinations in the 2.0< diameter ≤4.0 cm, and diameter >4.0 cm groups. Conclusions: With its non-invasive, safe, objective, and accurate advantages, dynamic AI can conduct real-time, multi-section, and multi-angle diagnosis. It can deeply analyze the subtle characteristics of large thyroid nodules and provide a more objective comprehensive assessment of thyroid nodules with a diameter >2 cm, enabling the precise diagnosis of the benignity or malignancy of these nodules. It promotes the homogenization of diagnosis and has important guiding significance for individualized diagnosis and treatment strategies, and the development of surgical resection ranges.