Predicting treatment outcome and prognosis in locally advanced rectal cancer using pretreatment MRI and pathology based machine learning models

Motivation: Monitoring and predicting treatment outcome in locally advanced rectal cancer is of clinical importance. Goal(s): To use machine learning models with the integration of complementary information from MRI-derived radiomics, deep learning, and pathology for the early prediction of treatment outcomes. Approach: Pretreatment T2 MRI, ADC and pathology-derived machine learning and deep learning models as well as a fusion model were developed for the prediction of pathologic complete response and local recurrence/distant metastasis. Results: The proposed models demonstrated high levels of performances for the prediction of early treatment outcomes. Impact: The results highlight the potential of machine learning models using MRI and pathology to enhance non-invasive prediction of rectal cancer treatment outcomes. This could lead to earlier, more tailored interventions and stimulate research into multimodal integration methods.