Intratumoral and peritumoral habitat imaging derived from multi-b-value Diffusion-Weighted imaging in rectal Cancer:...

To develop interpretable habitat imaging based on multi-b-value diffusion-weighted imaging (DWI) for characterizing tumor microenvironment features and predicting progression-free survival (PFS) in rectal cancer (RC).

This study enrolled pathologically confirmed patients with RC (n = 153) who underwent multi-b-value DWI before therapy. Intratumoral and peritumoral habitat subregions were coded using K-means clustering based on three maps (DDC_map, α_map and f_map). The associations between habitat features (fraction and Rad_score) and clinicopathological characteristics, as well as tumor microenvironment metrics (Glasgow Microenvironment Score (GMS)) were analysed using appropriate statistical methods. Univariable and multivariable Cox proportional hazards analysis were performed to assess the prognostic factors for RC. Four models (Clinical, Habitat, Habitat_Rad, and Nomogram) were developed for predicting 3-year PFS in RC. The performance of the four models was evaluated using Harrell's concordance index (C-index) and receiver operating characteristic (ROC) curves. The risk stratification of the Nomogram was investigated using Kaplan-Meier survival analysis and log-rank test.

Overall, 153 patients (n = 153; 59.25 ± 11.95 years) with RC were randomly divided into training (n = 107) and validation sets (n = 46). Intratumoral and peritumoral tissues were partitioned into three habitat subregions, respectively. Intratumoral habitat 3 (Intra-H3) fraction demonstrated a significant positive correlation with lymph node metastasis (p = 0.010), mrT-stage (p = 0.041) and GMS grade (p = 0.013), and served as an independent risk factor for PFS (HR = 1.957, 95%CI (1.209---3.167), p = 0.006). Peritumoral (0-3 mm) habitat 2 (Peri3-H2) fraction exhibited significant differences across Klintrup-Makinen (KM) grades (p=0.014). The Nomogram integrating habitat features demonstrated superior performance compared with conventional clinical model in the training (AUC = 0.900, 95% CI (0.835-0.966) VS 0.737, 95% CI (0.625-0.850), p = 0.002) and validation (AUC = 0.826, 95% CI (0.707-0.945) VS 0.697, 95% CI (0.524-0.871), p = 0.101) sets, respectively. The Kaplan-Meier curves of the Nomogram score showed good discriminative ability for 3-year PFS in RC.

Intra-H3 features reflect the tumor microenvironment and enhance predictive performance of 3-year PFS in RC. Integrating habitat features from multi-b-value DWI offers superior predictive performance, facilitating precise identification of high-risk patients.