T2WI-based intratumoral habitat and peritumoral radiomics to predict lymph node metastasis in cervical cancer: a multicenter study with biological validation

Abstract Background To develop and validate a T2WI-based nomogram integrating intratumoral habitat imaging, peritumoral radiomics, and clinical features for preoperative prediction of lymph node metastasis (LNM) in cervical cancer (CC), and to verify the biological plausibility of imaging habitats using an animal model. Methods This multicenter study included 766 CC patients (IB1-IIA2 and IIIC) from four centers, divided into training (n = 219), internal validation (n = 94), and two external test cohorts (n = 269, n = 184) between 2020 and 2024. Habitat imaging and peritumoral features were extracted, followed by machine‑learning model construction, evaluation, and SHAP interpretability. For biological validation, 12 nude mice bearing HeLa, SiHa, and U14 tumors underwent 9.4T MRI; imaging-pathology spatial registration was performed using 3D-printed molds, and habitat consistency was quantified by the Adjusted Rand Index (ARI) and intraclass correlation coefficient for area (ICC(area)). Results Clinical stage and tumor diameter ≥ 4 cm were independent risk factors for LNM (both P < 0.05). The combined model yielded AUCs of 0.933 (training), 0.898 (internal validation), and 0.858–0.863 (external tests). A nomogram with a Youden-derived cutoff of 0.32 demonstrated consistent sensitivity (0.818–0.913) and specificity (0.732–0.877) across cohorts, with good calibration (Hosmer-Lemeshow P = 0.183–0.561). Animal experiments revealed moderate spatial correspondence (ARI: 0.4–0.6) and high proportional consistency (ICC (area): 0.8–1.0) between imaging and pathological habitats. Conclusions This T2WI-based intratumoral habitat and peritumoral radiomics model provides accurate, economical preoperative LNM prediction for CC to guide personalized radiotherapy or surgery, with biological validity confirmed by animal experiments. Trial registration ChiCTR2400081731, Registered 11 March 2024, https//www.chictr.org.cn/showproj.html?proj=203940