A retrospective multicenter study with 313 patients (training cohort 183, test cohort 130) established an MRI model for preoperative differentiation of T1-2 from T3 stages in rectal cancer. The model grouped intratumoral radiomic features into three biotopes of the tumor microenvironment using K-means and extracted peritumoral features in the 1–3 mm range. Peritumoral (PERI2mm), biotope, clinical and nomogram models combining radiomics with clinical factors such as age, CEA, tumor length, circumferential growth and mrT stage were used. The nomogram achieved the highest AUC of 0.907 in the training and 0.881 in the test cohort, better than PERI2mm (AUC 0.817) or the clinical model (AUC 0.723). In the test cohort, it showed an accuracy of 0.827, a PR-AUC of 0.948 and an F1 score of 0.851. Calibration analysis confirmed agreement with pathological staging (Brier score 0.1253 in the test) and DCA showed clinical benefit. The nomogram improves the accuracy of T1-2/T3 differentiation using habitat, peritumoral radiomics and clinical biomarkers.