Increasing numbers of early rectal cancers are detected due to improved screening. MRI can assist in selecting lesions suitable for endoscopic resection[1]. At our institution, all suspicious rectal lesions undergo MRI assessment. We aimed to evaluate the accuracy of MRI in predicting the depth of submucosal invasion in early rectal cancers.

We conducted a retrospective study of T1/2 rectal cancers managed by endoscopic submucosal dissection between October 2019 and October 2025. Each MRI scan was independently reviewed by two radiologists, both blinded to endoscopic and histological findings. Radiologists assigned T-stage based on predefined MRI criteria: absence of submucosal invasion (T0); submucosa ≥1mm with intact muscularis propria (MP) (T1 sm1/2); submucosa <1mm but with ≥1mm intact MP (T1 sm3/early T2); or deep invasion characterised by <1mm intact MP or extension beyond it (deep T2/T3). Radiological staging was compared with final histology and Japan NBI expert team (JNET) or Kudo based optical assessment. Categorical variables were analysed using χ² testing, continuous variables with the Mann–Whitney U test, and inter-radiologist agreement using Cohen’s kappa (κ). 

Of the 21 cases identified, 13/21 (63%) had at least one technical issue. Motion artefact was the most common issue (8/13, 62%), followed by inadequate distension (3/13, 23%), absence of a true short-axis imaging (3/13, 23%), incomplete coverage (3/13, 23%) and faecal residue (1/13, 8%). However, this did not appear to affect diagnostic performance: MRI correctly staged 6/8 (75%) scans without technical issues and 11/13 (85%) with issues. Overall, one or both radiologists correctly staged 17/21 cases (81%). All radiologists correctly identified lesions with ≥1mm preserved MP which is a key determinant for endoscopic resectability. Staging concordance between radiologists was 13/21 (62%) consistent with moderate agreement (κ=0.48).

Optical assessment was then compared with MRI and histology. JNET 2B lesions were all T1 on histology except one T2 case, yet MRI overstaged 4/7 cases (57%), indicating a conservative bias for superficial disease. JNET 3 lesions showed strong alignment between all modalities, with MRI correctly staging 5/6 cases (83%). Kudo IV lesions were heterogenous, including overstaging of one Tis lesion. Among Kudo V lesions, histological depth varied; MRI correctly staged 5/6 cases (83%) and overstaged one case with no understaging. Overall, optical assessment correlated more closely with histology for superficial disease, but MRI performed best for lesions with deeper invasion. MRI tended to overstage disease compared with final histology (9/21; 43%), with no instances of understaging. 

Table 1 summarises clinical and procedural characteristics stratified by MRI accuracy. There were no identifiable predictors of correct staging (all p>0.05).

MRI demonstrated high accuracy in identifying early rectal cancers suitable for endoscopic resection, with excellent sensitivity for detecting preserved MP. Accuracy was unaffected by frequent technical issues, although inter-radiologist agreement was moderate and MRI consistently overstaged disease. Optical assessment corresponded more closely to histology for superficial disease, whereas MRI performed best for deeper lesions. MRI is a safe and conservative staging tool, with scope for improvement through targeted MDT feedback.