Radiomics-based Machine Learning Approach to Predict Chemotherapy Responses in Colorectal Liver Metastases

• DOI: 10.23922/jarc.2024-077
• Publication Date: 2025-01-24T22:13:01Z
• Authors: Yuji Miyamoto; Takeshi Nakaura; Mayuko Ohuchi; Katsuhiro Ogawa; Rikako Kato; Yuto Maeda; Kojiro Eto; Masaaki Iwatsuki; Yoshifumi Baba; Toshinori Hirai; Hideo Baba
• Keywords: machine learning; colorectal cancer; RC799-869; Original Research Article; Diseases of the digestive system. Gastroenterology; chemotherapy; liver metastases; ct texture analysis

This study explored the clinical utility of CT radiomics-driven machine learning as a predictive marker for chemotherapy response in colorectal liver metastasis (CRLM) patients. We included 150 CRLM patients who underwent first-line doublet chemotherapy, dividing them into training cohort (n=112) and test (n=38). manually delineated three-dimensional tumor volumes, selecting largest measurement, using pretreatment portal-phase images extracted 107 radiomics features. Treatment was classified responder (complete or partial response) non-responder (stable progressive disease), based on best overall according to RECIST criteria, version 1.1. Employing Random Forest Boruta algorithms, we identified significant features responder-non-responder differentiation. Radiomics signatures were developed validated five-fold cross-validation, performance assessed area under curve (AUC). Among patients, 91 (61%) responders 59 (39%) non-responders. Variable selection with revealed three key parameters ("DependenceVariance," "ClusterShade," "RunVariance"). In cohort, individual texture parameter AUCs ranged from 0.4 0.65, while analysis incorporating all valid exhibited significantly higher AUC 0.94 (p<0.01). The validation also demonstrated strong accuracy, an 0.87 treatment response. highlights potential predicting responses among