Combined Lipid Metabolism Indicators and CT Deep Learning Model to Predict Rectal Cancer Prognosis

Abstract

Objective Colorectal cancer (CRC) is a common malignant tumor, and its prognosis is closely associated with abnormalities in lipid metabolism; meanwhile, CT imaging-based radiomics also demonstrates significant advantages in the evaluation of CRC. This study aims to investigate the application value of lipid metabolism indicators and CT deep learning models in prognostic assessment of colorectal cancer. Materials and Methods We retrospectively collected clinical data, lipid metabolism indicator data, and CT imaging data of patients diagnosed with rectal cancer by pathology at Gansu Provincial People's Hospital from January 2019 to May 2021. First, we analyzed the patients' lipid metabolism indicators, including serum lipid levels and metabolic-related biomarkers. Next, we delineated regions of interest (ROIs) from the patients' CT images. After delineation, we used a cropping tool to extract the tumor's maximum ROI cross-section for deep learning model training. We trained deep learning models using the ResNet and DenseNet series, plotted ROC curves, and calculated the AUC (area under the curve) to evaluate the performance of the predictive models. We also plotted calibration curves and DCA (decision curve analysis) curves to assess the model's fit and clinical value, and drew KM curves to evaluate its relationship with prognosis. Finally, we used a nomogram to integrate lipid metabolism indicators and the CT deep learning model at the outcome level, establishing a combined model for predicting rectal cancer prognosis. Results The predictive model based on lipid metabolism indicators achieved an AUC of 0.700. The CT-based deep learning model achieved an AUC of 0.746 on the test set. The combined model achieved an AUC of 0.863. Calibration curve analysis demonstrated superior accuracy of the combined model, while decision curve analysis (DCA) indicated better clinical utility. Kaplan-Meier curves showed that risk stratification based on the combined model accurately assessed patients' prognosis. Conclusion The combined model based on lipid metabolism indicators and CT deep learning has certain predictive value for the prognosis of rectal cancer.