An MRI-based habitat radiomics nomogram for preoperative prediction of lymphovascular invasion in rectal cancer patients.

OBJECTIVE: This study seeks to assess the potential of using an MRI-based habitat radiomics nomogram to predict lymphovascular invasion (LVI) preoperatively in rectal cancer (RC) patients, while also comparing its performance to traditional intratumoral radiomics models. METHODS: This study retrospectively included 197 RC patients diagnosed with LVI, who were randomly assigned to a training cohort (n = 139) and a validation cohort (n = 58). The K-means algorithm was used to perform clustering on the arterial phase of contrast-enhanced MRI images. Following feature extraction and selection, separate models were developed for intratumoral radiomics and habitat analysis to preoperatively predict LVI status in RC patients. A nomogram was created by combining the intratumoral radiomics model, habitat radiomics model, and clinical model. The performance and advantages of the model were evaluated through the area under the receiver operating characteristic curve (AUC), calibration curve and decision curve analysis (DCA). RESULTS: Three habitats were clustered, and in the validation cohort, the habitat radiomics model demonstrated a higher AUC of 0.783, outperforming the intratumoral radiomics model, which had an AUC of 0.743. The nomogram, combining intratumoral radiomics features, habitat characteristics, and the clinical independent predictor (mrN stage), demonstrated the highest AUC of 0.833. CONCLUSION: The findings indicate that the model utilizing habitat characteristics demonstrates superior performance compared to the traditional intratumoral radiomics model. The nomogram, which incorporates intratumoral and habitat radiomic features along with the independent clinical predictor, demonstrated the highest accuracy for preoperative prediction of LVI in RC, representing a significant advancement in model performance.