An Artificial Intelligence–Driven Preoperative Radiomic Subtype for Predicting the Prognosis and Treatment Response of Patients with Papillary Thyroid Carcinoma

Abstract

PURPOSE: From 8% to 28% of patients with papillary thyroid carcinoma (PTC) experience recurrence, complicating risk stratification and treatment. We previously identified an inflammatory molecular subtype of PTC associated with poor prognosis. Based on this subtype, we aimed to develop and validate a noninvasive radiomic signature to predict prognosis and treatment response in patients with PTC. EXPERIMENTAL DESIGN: We collected preoperative ultrasound images from two large independent centers (n = 2,506) to develop and validate a deep learning radiomics signature of inflammation (DLRI) for predicting the inflammatory subtype of PTC, including its correlation with prognosis and anti-inflammatory traditional Chinese medicine (TCM) treatment. Training set 1 (n = 64) and internal validation set 2 (n = 1,108) were from Tianjin Medical University Cancer Institute and Hospital. External validation sets 1 (n = 76) and 2 (n = 1,258) were from Fudan University Shanghai Cancer Center. RESULTS: We developed a DLRI to accurately predict PTC's inflammatory subtype (AUC = 0.97 in training set 1 and AUC = 0.82 in external validation set 1). High-risk DLRI was significantly associated with poor disease-free survival in the first cohort [HR = 16.49, 95% confidence interval (CI), 7.92-34.35, P < 0.001] and second cohort (HR = 5.42, 95% CI, 3.67-8.02, P < 0.001). The DLRI independently predicted disease-free survival, irrespective of clinicopathologic variables (P < 0.001 for all). Furthermore, patients with high-risk DLRI were likely to benefit from anti-inflammatory TCM treatment (HR = 0.19, 95% CI, 0.06-0.55, P = 0.002), whereas those with low-risk DLRI did not. CONCLUSIONS: DLRI is a reliable noninvasive tool for evaluating prognosis and guiding anti-inflammatory TCM treatment in patients with PTC. Prospective studies are needed to confirm these findings.

Ähnliche Arbeiten