Ultrasound deep learning features for predicting cervical lymph node metastasis in papillary thyroid carcinoma.

Background: The purpose of this study was to create and validate a clinical deep learning radiomics (DLR) model for determining cervical lymph node metastasis (CLNM) status in patients with papillary thyroid cancer (PTC). Methods: A total of 205 eligible patients with PTC who underwent preoperative thyroid ultrasonography (US) between January 2015 and April 2020 were retrospectively enrolled. The training cohort consisted of 143 patients, while the validation cohort included 62 patients. A DLR model was built using deep learning features, a clinical model was built using clinical parameters, and a Cli-DLR model was built using both DLR features and clinical factors. Results: In the validation cohort, the Cli-DLR model performed well, with an AUC of 0.80. The Cli-DLR model performed well in terms of precision-recall and calibration curve analysis. Furthermore, the Cli-DLR model outperformed experienced radiologists. This approach has the potential to help guide optimal CLNM management in PTC patients, particularly by preventing overtreatment. Conclusions: In conclusion, we developed a Cli-DLR model to assess the cervical lymph node status of PTC patients prior to surgery. The Cli-DLR model outperformed radiologists and the clinical model. Consequently, this model can provide a possible noninvasive method for detecting CLNM and aid in clinical decision-making due to its favorable specificity and sensitivity. High-level evidence for clinical use in later studies is anticipated to be obtained through prospective multicenter validation.