Fusion model predicts KRAS mutations in rectal cancer effectively

A study evaluated deep learning, radiomics, and fusion models to predict KRAS mutations in rectal cancer using endorectal ultrasound images from 304 patients. Among the models, the feature-based fusion model (dlrexpand10_fb) achieved the highest area under the receiver operating characteristic curve (AUC) of 0.896, indicating superior predictive performance. Additionally, incorporating peritumoral regions significantly improved the outcomes, highlighting the deep learning feature (dl_1) as crucial in determining mutation probabilities.