Key points are not available for this paper at this time.
This study aimed to develop and validate a fully automated magnetic resonance imaging (MRI) deep learning (DL) framework for primary tumor detection and prediction of lymph node metastasis (LNM) in early-stage OTSCC. A total of 348 patients from two centers were retrospectively enrolled and were divided into a training cohort (n = 163), a validation cohort (n = 65), an internal testing cohort (n = 84) and an external testing cohort (n = 36). The You Only Look Once version 8 (YOLOv8) network was used as the backbone for automated primary tumor detection. Two distinct image encoders were employed to construct the DL models for predicting LNM. The automated DL signature was combined with significant clinical factors to develop the automated DL nomogram. Predictive performance was assessed using the area under the curve (AUC). The prognostic value of automated DL signature was determined by Cox regression analysis. The YOLOv8 model demonstrated effective primary tumor detection ability on MRI, achieving mean average precisions (mAPs) of 0.689–0.973 across the validation, internal, and external testing cohorts. The automated DL nomogram, which combined the automated DL signature and clinical T stage, achieved the best performance (AUCs: 0.871, 0.776, 0.909) and significantly outperformed clinical T stage alone (AUCs: 0.653, 0.611, 0.685, all P < 0.01) across the same cohorts. DL signature–predicted positive LNM was significantly associated with worse disease-free survival following surgical treatment (hazard ratio, 1.82; 95% CI: 1.04, 3.19; P = 0.037). The MRI-based DL framework facilitates predictions of LNM and recurrence risk, thereby assisting in decision-making for patients with early-stage OTSCC.
Ren et al. (Mon,) studied this question.