EP363 - MRI Accuracy in Rectal Cancer Staging: Is it as Good as We Think

Purpose/Background: Colorectal cancer is a leading cause of cancer-related death in the US, with diagnoses increasing each year. Rectal protocol MRI is the standard imaging for local staging, including assessing depth of invasion (T stage), nodal involvement (N stage), circumferential resection margin (CRM), and extramural venous invasion (EMVI). Literature on MRI accuracy is mixed, leading to potential risks of over or under-treatment. This study assessed MRI accuracy at the University of Vermont Medical Center (UVMMC).

Methods/Interventions: We conducted a retrospective chart review as a quality improvement initiative at UVMMC. Rectal cancer patients were divided into three groups.

Group 1 included those proceeding directly to surgery without neoadjuvant therapy; preoperative MRI was compared with surgical pathology which we considered the gold standard for T, N, EMVI, and CRM staging.

Group 2 included patients who received total neoadjuvant therapy (TNT) followed by surgery; again, preoperative MRI was compared with the surgical pathology. In this group radiologic tumor response was defined as complete, near complete or incomplete. For analytic purposes group 2 was subdivided into 2a: complete clinical response (cCR) followed by surgery, and 2b: incomplete clinical response. MRI was compared with pathology for primary tumor response, N stage, EMVI, and CRM.

Group 3 included patients treated with neoadjuvant chemoradiation without surgery, either as planned non-operative management, or because other clinical factors precluded surgery. Treatment response was evaluated by comparing post-therapy flexible sigmoidoscopy (flex sig) with the chronologically closest MRI.

Note: Several patients in this study were also included in the OPRA trial.

Results/Outcomes: This study included 122 patients from October 2015 to August 2025. In Group 1 (27 patients), MRI accuracy was: T stage 67% (18/27, 95% CI 46%-83%), N stage 63% (17/27, 95% CI 42%-81%), EMVI 54% (14/26, 95% CI 33%-73%), and CRM 93% (25/27, 95% CI 76%-99%).

Group 2 (33 patients; 2a: 7, 2b: 26), In group 2a primary tumor treatment response accuracy was 71% (5/7, 95% CI 29%-96%), N stage 100% (7/7, 95% CI 59%-100%), EMVI 86% (6/7, 95% CI 42%-100%), and CRM 86% (6/7, 95% CI 42%-100%).

In group 2b primary tumor treatment response accuracy was 100% (25/25, 95% CI 87%-100%), N stage 65% (15/23, 95% CI 43%-84%), EMVI 69% (18/26, 95% CI 48%-86%), and CRM 73% (19/26, 95% CI 52%-88%).

In Group 3 (62 patients), flex sig correctly identified complete or incomplete clinical response in 93% (40/43), compared with MRI 56% (24/40). When MRI and flex sig agreed, the combined accuracy was 93% (27/29).

Conclusion/Discussion: At UVMMC, an academic medical center with dedicated radiologists specializing in rectal protocol MRI, pretreatment staging MRI most accurately assessed CRM, but was less accurate for tumor and nodal staging, and EMVI. Accuracy declined further after neoadjuvant therapy, especially for primary tumor response assessment. Flex sig significantly outperformed MRI in this setting, underscoring its value in non-operative management. MRI remains essential for baseline staging, particularly for CRM, but shows limited accuracy for T and N stage and EMVI. Following chemoXRT flex sig more reliably predicts treatment response, particularly cCR, supporting its role in non-operative management. Future directions include evaluating non generative artificial intelligence for its ability to increase MRI accuracy. We are currently expanding our project to evaluate this technology.