An Integrative Clinical and Intra- and Peritumoral MRI Radiomics Nomogram for the Preoperative Prediction of Lymphovascular Invasion in Rectal Cancer.

Abstract

Rationale and objectives: Accurately and noninvasively predicting lymphovascular invasion (LVI) is critical for the prognosis of patients with rectal cancer (RC). The objective of this study was to create a nomogram model that combines clinical features with MRI-based radiomic characteristics of both intratumoral and peritumoral regions to predict LVI in patients with resectable rectal cancer.

Method: This study retrospectively included 149 RC patients diagnosed with LVI, who were randomly assigned to a training cohort (n=104) and a testing cohort (n=45). Radiomics features were derived from intratumoral and peritumoral areas using different expansion dimensions (3 and 5 mm) in T2-weighted imaging (T2WI) and Diffusion-Weighted Imaging (DWI). A nomogram was created by combining the optimal radiomics model with the most predictive clinical factors to enhance the LVI prediction.

Results: In the validation cohort, the radiomics models using 3 mm and 5 mm peritumoral regions in T2WI achieved AUC values of 0.786 and 0.675, respectively, surpassing the performance of models based on DWI. In both T2WI and DWI, the 3 mm peritumoral model outperformed the 5 mm model in predictive accuracy. Therefore, the combined radiomics model integrating intratumoral and the 3 mm peritumoral regions in T2WI was identified as the optimal radiomics model, achieving an AUC of 0.913. The decision and calibration curves showed that radiomics models incorporating both intratumoral and peritumoral regions outperformed traditional approaches. A nomogram was created by combining a clinical model that incorporates gender and mrN stage with the optional radiomics model, aiming to predict LVI in patients with RC.

Conclusion: The radiomics model based on the 3 mm peritumoral region in T2WI demonstrated greater precision and sensitivity in identifying LVI. The radiomics model, which combined features from both intratumoral and peritumoral regions, exhibited superior performance compared to models based solely on either intratumoral or peritumoral attributes. The optimal combination was the integration of intratumoral features with the 3 mm peritumoral region in T2WI. A nomogram integrating radiomics features from intratumoral and peritumoral regions with clinical parameters offers valuable support for the preoperative diagnosis of LVI in RC, demonstrating significant clinical utility.