Progress of MRI‑based radiomics and deep learning for predicting the prognosis of locally advanced rectal cancer (Review)

Rectal cancer (RC) ranks among the most common malignant tumors worldwide, with ~30% of patients presenting at a locally advanced stage at the time of diagnosis. The standard treatment for locally advanced RC (LARC) combines neoadjuvant chemoradiotherapy (nCRT) with total mesorectal excision. While this treatment paradigm has been effective in reducing the local recurrence rate, its efficacy in enhancing overall survival and disease-free survival is still limited. Consequently, the identification of adverse prognostic factors in patients with LARC is crucial for improving long-term survival outcomes. Traditional imaging methods offer limited predictive power for early diagnosis, treatment efficacy assessment and prognosis in LARC. Recent advancements in MRI-based radiomics and deep learning (DL), leveraging high-dimensional feature extraction and nonlinear modeling, have markedly enhanced prognostic predictive performance. Single-sequence MRI radiomics models derived from pre-nCRT imaging demonstrate areas under the curve (AUC) of 0.79-0.87 for predicting local recurrence and distant metastasis. Multiparametric radiomic models further improve predictive accuracy, achieving AUCs of 0.81-0.83. Delta radiomics, which captures temporal-spatial heterogeneity evolution in tumors during therapy, elevates AUC performance to 0.77-0.89. Notably, DL-based models exhibit superior and more stable predictive capabilities, with concordance indices (C-indices) ranging from 0.82 to 0.94. This paper reviews recent progress in MRI-based radiomics and DL for predicting the prognosis of patients with LARC subjected to nCRT.