MRI Habitat Analysis for Preoperative Prediction of Perineural Invasion and Prognostic Stratification in Rectal Cancer.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiology, Tongde Hospital of Zhejiang Province, Hangzhou, Zhejiang Province, China.
- Zhejiang Academy of Traditional Chinese Medicine, Hangzhou, Zhejiang Province, China.
- The Integrated Traditional Chinese and Western Medicine School of Clinical Medicine, Zhejiang Chinese Medical University, Hangzhou, Zhejiang Province, China.
- Department of Pathology, Tongde Hospital of Zhejiang Province, Hangzhou, Zhejiang Province, China.
- Department of Radiology, Putuo People's Hospital, School of Medicine, Tongji University, Shanghai, China.
Abstract
Accurate preoperative assessment of perineural invasion (PNI) remains challenging in rectal cancer. To develop assessment models based on preoperative multiparametric MRI (mpMRI) habitat analysis for evaluating PNI status and to explore their prognostic value. Retrospective. Six hundred and twenty-one rectal cancer patients were enrolled from two centers, divided into a training set (n = 330; 65.8 ± 11.22 years; 215 males), an internal validation set (in-vad, n = 152; 67.85 ± 12.43 years; 105 males), and an external validation set (ex-vad, n = 139; 62.82 ± 11.79 years; 96 males). 1.5T, 3T, T2-weighted imaging using turbo spin-echo sequence, diffusion-weighted imaging using echo planar imaging, and contrast-enhanced T1-weighted imaging using 3D spoiled gradient echo sequence. Tumor voxels were partitioned into subregions using k-means clustering, and habitat-based submodels were developed with deep learning. The Boruta algorithm combined with univariate and multivariate analyses identified key variables. Student's t test, Mann-Whitney U test, chi-square test, Boruta analysis, and DeLong's test. Significance was defined as p < 0.05. A clinical model was constructed from selected significant variables, and a nomogram integrating the clinical model with habitat-based submodels was subsequently developed. Tumors were divided into three imaging-derived subregions, generating three habitat submodels. Habitat 1, 2, 3, mrN, and mrEMVI were independent PNI variables. The nomogram exhibited the highest performance, with area under the curve (AUC) values of 0.967 (95% confidence interval [CI], 0.950-0.983), 0.965 (0.941-0.990), and 0.977 (0.949-1.000) in the training, in-vad, and ex-vad sets, respectively. Kaplan-Meier analysis further confirmed its effective stratification of 3-year disease-free survival. The MRI-based habitat analysis model and the derived nomogram demonstrate high predictive value for preoperative assessment of PNI in rectal cancer. The nomogram also shows promising capability for prognostic risk stratification. 3.