Predicting treatment response for the safe non-operative management of patients with rectal cancer using an MRI-based deep-learning model.

作者：Heather M. Selby, Charles Liu, Vipul Sheth, Sandy Napel, Todd H. Wagner, Arden M. Morris · 发表于：Journal of Clinical Oncology · 年份：2023 · DOI：10.1200/jco.2023.41.16_suppl.e15648 · 研究领域：Radiomics and Machine Learning in Medical Imaging、Colorectal Cancer Surgical Treatments、MRI in cancer diagnosis

e15648 Background: Neoadjuvant chemoradiotherapy (CRT) followed by total mesorectal excision (TME) surgery has been the standard treatment for patients with locally advanced rectal cancer. Surgery, while effective in treating rectal cancer, can lead to significant long-term morbidity, such as bowel, urinary, and sexual dysfunction, and potential dependence on a colostomy for evacuation. A pathologic complete response (pCR) to neoadjuvant CRT suggests that TME surgery may not be necessary, as the tumor may have been effectively treated with CRT alone. However, the only definitive determination of pCR is based on a pathologic exam, which requires a surgical specimen. By comparing pre- and post-CRT images, clinical complete response (cCR) is a surrogate for pCR that does not require surgery. However, cCR is a poor proxy for pCR, with sensitivity and positive predictive values of 34.1% and 30.6% respectively 1 . Purpose: Our aim is to develop a multiparametric longitudinal MRI-based 3D deep-learning model to predict pCR in patients with locally advanced rectal cancer. Methods: We identified 360 patients with rectal cancer who had neoadjuvant CRT followed by surgical resection between January 2012 and December 2021 in an academic medical center. We have classified patients into two groups based on the presence or absence of pCR in their resected tumor specimen. We will train and validate a multiparametric longitudinal MRI-based 3D deep-learning model to predict pCR. Our model will...