Habitat-based amide proton transfer-weighted MRI model for predicting BRAF mutation and prognostic stratification in...

This study aims to investigate the efficacy of a novel habitat-based histogram features derived from amide proton transfer-weighted (APTw) MRI for predicting BRAF mutation status in RC, compared and combined with diffusion-weighted imaging (DWI), and the potential for prognostic stratification.

This study prospectively enrolled 269 patients with RC from June 2021 to August 2025, divided into a training set (n=188) and a testing set (n= 81) using a fixed random seed. According to BRAF status, patients were categorized into a wild-type group (n=214) and a mutant group (n=55). K-means clustering was applied to partition the tumors into 7 sub-regions, from which corresponding histogram features of APTw and apparent diffusion coefficient (ADC) maps were extracted. The extracted features were subsequently filtered using stepwise regression. The predictive performance of five models-namely, the habitat-based APTw model, habitat-based ADC model, habitat-based APTw+ADC combined model, clinical factors model, and the comprehensive model integrating all features-was evaluated using receiver operating characteristic (ROC) curve analysis and decision curve analysis (DCA). Additionally, Kaplan-Meier survival curves, constructed based on the combined nomogram, were used to assess 2-year disease-free survival (DFS) in the entire cohort.

The K-means model with K = 7 demonstrated optimal predictive power for BRAF mutation, outperforming models with K = 5 or 6, with the mutant group showing significant differences in APT/ADC histogram features in specific sub-regions compared to the wild-type group. The nomogram integrating APTw/ADC subregion histogram features and the clinical factors showed superior discrimination, achieving an AUC of 0.83 (95% CI: 0.707-0.952), which was significantly higher than that of the habitat-based APTw model (0.733), habitat-based ADC model (0.71), or clinical factors model (0.661) (all P < 0.05) in the testing set. Furthermore, patients stratified into the mutant group by the nomogram had a significantly worse 2-year disease-free survival (DFS) than those in the wild-type group (P < 0.05).

Habitat-based APT/ADC histogram features combined with clinical model showed good performance in predicting BRAF mutation and DFS of RC patients, which could provide valuable information for individualized treatment.