Machine learning-based radiomic models for predicting metachronous liver metastases in colorectal cancer patients: a multimodal study

Abstract

Abstract Purpose The purpose of this study was to investigate whether a multimodal radiomic model powered by machine learning (ML) can accurately predict the occurrence of metachronous liver metastases (MLM) in patients with colorectal cancer (CRC). Patients and methods: A total of 157 consecutive patients with CRC between 2010 and 2020 were retrospectively included. Out of these patients, 67 experienced liver metastases within 2 years of treatment, while the remaining patients did not. Radiomic features were extracted from annotated MR images of the tumor and portal venous phase CT images of the liver for each patient. Following that, ML-based radiomic models were then developed and integrated with the clinical features for MLM prediction by employing LASSO and RF algorithms. The performance of the model was evaluated using the ROC curve, while the clinical utility was measured using the DCA curve. Results A total of 922 and 1216 radiomic features were extracted from the MRI and CT images of each patient, quantifying the intensity, shape, orientation, and texture of the tumor and liver, respectively. The mean area under the curves(AUCs) for predicting metachronous liver metastases were 0.80, 0.68, and 0.82 for the CT, MRI, and Merged models, respectively. For the Clinical and Clinical-Merged models, the AUCs were 0.64 and 0.72, respectively. There was no significant difference between the CT model and the Merged model (p < 0.05). Conclusion Our preliminary results demonstrate the utility of ML-based radiomic models in predicting MLM in patients with CRC. However, further investigation is required to explore the potential of multimodal fusion models, as they offered only minimal improvement in diagnostic performance.