Deep learning model effectively predicts mismatch repair deficiency in colorectal cancer

A new deep learning model demonstrated strong predictive capabilities for mismatch repair deficiency in colorectal cancer. An area under the receiver operating characteristic (AUROC) score of 0.948 was achieved during testing, while independent validation yielded an AUROC of 0.807. Notably, the model achieved a negative predictive value of 94.2%. In a prospective trial, it accurately identified 29 out of 33 deficient tumors, suggesting its potential as an automatic screening tool for clinicians.