Macro Habitat-Based T2-Weighted MRI Radiomics and Deep Learning Fusion for Predicting Treatment Response and Prognosis After Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer

Comprehensive Summary

This study develops a method to predict how patients with locally advanced rectal cancer respond to neoadjuvant chemoradiotherapy using pretreatment T2 weighted MRI. Because treatment response varies widely, the authors aimed to improve prediction before therapy begins. They analyzed 434 patients from multiple centers and introduced a macro habitat approach that divides the tumor into multiple subregions based on imaging patterns while also including surrounding tissue. From these regions, they extracted radiomic features that quantify texture and structure along with deep learning features from a 3D neural network. These features were combined and used in machine learning models to predict tumor regression grade and progression free survival. The results showed that models using only tumor or surrounding tissue performed modestly, while the combined macro habitat model achieved much higher accuracy, with strong performance in external validation. The model also produced risk scores that were significantly associated with patient outcomes, distinguishing between better and worse prognoses. This suggests that integrating tumor heterogeneity and the surrounding microenvironment improves prediction and could support more personalized treatment decisions, such as selecting patients for organ preserving strategies or more aggressive therapy. Limitations include the retrospective design, use of a single MRI sequence, and variability across centers, so further prospective validation is needed.

Outcomes and Implications

The clinical implications of this study are significant because it offers a noninvasive and more accurate way to predict treatment response and prognosis in patients with locally advanced rectal cancer before therapy begins. By identifying which patients are likely to respond well to neoadjuvant chemoradiotherapy, clinicians could select candidates for organ preserving strategies such as watch and wait, potentially avoiding surgery and its associated complications. At the same time, patients predicted to respond poorly could be directed toward more intensive or alternative treatment approaches earlier in their care. The model’s ability to also predict progression free survival means it could help guide follow up intensity and long term management decisions. Overall, this approach supports more personalized treatment planning, improves risk stratification, and may lead to better outcomes while reducing unnecessary interventions.