Prediction of Local Tumor Progression After Microwave Ablation in Colorectal Carcinoma Liver Metastases Patients by MRI Radiomics and Clinical Characteristics-Based Combined Model: Preliminary Results.

ABSTRACT

PURPOSE: To investigate the predictability of local tumor progression (LTP) after microwave ablation (MWA) in colorectal carcinoma liver metastases (CRLM) patients by magnetic resonance imaging (MRI) radiomics and clinical characteristics-based combined model. MATERIALS AND METHODS: Forty-two consecutive CRLM patients (67 tumors) with post-MWA complete response at 1st month MRI were included in this retrospective study. One hundred and eleven radiomics features were extracted for each tumor and for each phase by manual segmentation from pre-treatment MRI T2 fat-suppressed (Phase 2) and early arterial phase T1 fat-suppressed sequences (Phase 1). A clinical model was constructed using clinical data, two combined models were created with feature reduction and machine learning by combining clinical data and Phase 2 and Phase 1 radiomics features. The predicting performance for LTP development was investigated. RESULTS: LTP developed in 7 patients (16.6%) and 11 tumors (16.4%). In the clinical model, the presence of extrahepatic metastases before MWA was associated with a high probability of LTP (p < 0.001). The pre-treatment levels of carbohydrate antigen 19-9 and carcinoembryonic antigen were higher in the LTP group (p = 0.010, p = 0.020, respectively). Patients with LTP had statistically significantly higher radiomics scores in both phases (p < 0.001 for Phase 2 and p = 0.001 for Phase 1). The classification performance of the combined model 2, created by using clinical data and Phase 2-based radiomics features, achieved the highest discriminative performance in predicting LTP (p = 0,014; the area under curve (AUC) value 0.981 (95% CI 0.948-0.990). The combined model 1, created using clinical data and Phase 1-based radiomics features (AUC value 0,927 (95% CI 0.860-0.993, p < 0.001)) and the clinical model alone [AUC value of 0.887 (95% CI 0.807-0.967, p < 0.001)] had similar performance. CONCLUSION: Combined models based on clinical data and radiomics features obtained from T2 fat-suppressed and early arterial-phase T1 fat-suppressed MRI are valuable markers in predicting LTP after MWA in CRLM patients. Large-scale studies with internal and external validations are needed to come to a firm conclusion on the predictability of radiomics models in CRLM patients.