Machine learning enhances assessment of colorectal liver metastases response

A machine learning model utilizing CT-based radiomics significantly surpassed traditional radiologist assessments in estimating pathologic response of colorectal liver metastases after neoadjuvant therapy. In a study involving 85 patients, the model achieved an area under the curve (AUC) of 0.87, contrasting sharply with the AUCs of 0.53 for RECIST assessments and 0.56 for morphologic evaluation. Subjective imaging characteristics showed no correlation with pathologic response, emphasizing the effectiveness of radiomics.