Abstract 5783: CT radiomics-based machine learning models for diagnosing para-aortic lymph node metastases in colorectal cancer

Abstract Background: Para-aortic lymph node (PALN) metastasis is a rare but clinically challenging manifestation in colorectal cancer (CRC). Its management is complicated by the high risks associated with radical PALN dissection and the limited diagnostic accuracy of preoperative imaging. This study aims to develop CT radiomics-based machine learning models to predict PALN metastasis in CRC patients. Methods: A retrospective analysis was conducted on 116 CRC patients who were enrolled between December 2012 and December 2019 and underwent PALN dissection and preoperative enhanced CT imaging within two weeks before surgery. CT images were semi-automatically segmented using 3D Slicer to delineate PALNs as regions of interest, from which a series of shape and texture radiomics features were extracted. Key features were selected using least absolute shrinkage and selection operator (LASSO) regression with cross-validation. Machine learning models, including logistic regression, support vector machine (SVM), extreme gradient boosting (XGBoost), and random forest, were constructed and cross-validated to assess the predictive value of these radiomics features for PALN metastasis. Results: A total of 1,037 CT radiomics features were extracted from the regions of interest, with six features selected based on their recurrence in over 200 instances. All four machine learning models performed well, with the random forest model exhibiting the most balanced performance: an average accuracy of 0.836, average precision of 0.842, average recall of 0.978, average F1 score of 0.905, and average AUC of 0.809. Compared to traditional diagnostic criteria based on lymph node short-axis diameter, the random forest model significantly improved diagnostic performance, with a net reclassification improvement (NRI) of 0.789 (95% CI, 0.378-1.199, P< 0.001) and an integrated discrimination improvement (IDI) of 0.116 (95% CI, 0.024-0.208, P= 0.014). Conclusion: CT radiomics-based machine learning models provide a robust, non-invasive tool for predicting PALN metastasis in CRC, offering superior diagnostic accuracy over traditional imaging metrics. Keywords: CT radiomics, Machine learning model, Para-aortic lymph node, Lymph node metastasis, Colorectal cancer, Prediction. Funding: Supported by National Key R&D Program of China (No. 2023YFC2507400), grants from the Guangdong Natural Science Foundation (No. 2023A1515011292), the Open Project of Guangdong Provincial Key Laboratory of Artificial Intelligence in Medical Image Analysis and Application (No. 2022B1212010011), National Key Clinical Discipline, The program of Guangdong Provincial Clinical Research Center for Digestive Diseases (No. 2020B1111170004). Yingqian Zhou, Ximin Pan and Qiongwei Tang contributed equally to this study. Corresponding authors: Yufeng Chen, [email protected]; Peiyi Xie, [email protected]. Citation Format: Yingqian Zhou, Ximin Pan, Qiongwei Tang, Zhaoliang Yu, Zerong Cai, Peiyi Xie, Yufeng Chen. CT radiomics-based machine learning models for diagnosing para-aortic lymph node metastases in colorectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5783.