Combined apparent diffusion coefficient histogram and MRI-based fractal analysis for predicting lymph node metastasis in esophageal squamous cell carcinoma

To investigate the feasibility of whole-tumor apparent diffusion coefficient (ADC) histogram analysis based on paired b-value ADC maps combined with fractal and morphological parameters of esophageal squamous cell carcinoma (ESCC) for predicting lymph node metastasis (LNM), 117 prospective ESCC patients from two centers were assigned to the training ( n = 63), internal ( n = 27) and external validation ( n = 27) cohorts. They underwent preoperative T 1 WI and T 2 WI, contrast-enhanced T 1 WI (T 1 CE), and DWI with five b-value combinations including 0 and 800 s/mm 2 . Whole-tumor ADC histogram features from DWI, fractal parameters at all sequences, and gross tumor volume (GTV) at T 1 CE were extracted. In the training cohort, five paired b-value based ADC histogram models, one fractal-dimension (FD) model, and one combined model were developed. In the three cohorts, models’ performance was assessed using area under the ROC curve (AUC), calibration curves, and Brier scores. Our combined model, incorporating the 25th percentile of ADC 0−800 with FD and natural-log-transformed GTV, yielded the highest AUCs (0.846–0.931 across the three cohorts). Calibration analysis showed reasonable overall agreement but slight deviations at higher predicted probabilities, and the combined model yielded Brier scores of 0.093 to 0.176 across cohorts. In conclusion, our combined model may provide a promising noninvasive approach for predicting LNM.