OBJECTIVE: The objective of this study is to find a contrast-enhanced CT-radiomic signature to predict clinical incomplete response in patients affected by hepatocellular carcinoma who underwent locoregional treatments. PATIENTS AND METHODS: 190 patients affected by hepatocellular carcinoma treated using focal therapies (radiofrequency or microwave ablation) from September 2018 to October 2020 were retrospectively enrolled. Treatment response was evaluated on a per-target-nodule basis on the 6-months follow-up contrast-enhanced CT or MR imaging using the mRECIST criteria. Radiomics analysis was performed using an in-house developed open-source R library. Wilcoxon-Mann-Whitney test was applied for univariate analysis; features with a p-value lower than 0.05 were selected. Pearson correlation was applied to discard highly correlated features (cut-off=0.9). The remaining features were included in a logistic regression model and receiver operating characteristic curves; sensitivity, specificity, positive and negative predictive value were also computed. The model was validated performing 2000 bootstrap resampling. RESULTS: 56 treated lesions from 42 patients were selected. Treatment responses were: complete response for 26 lesions (46.4%), 18 partial responses (32.1%), 10 stable diseases (17.9%), 2 progression diseases (3.6%). Area-Under-Curve value was 0.667 (95% CI: 0.527-0.806); accuracy, sensitivity, specificity, positive and negative predictive values were respectively 0.66, 0.85, 0.50, 0.59 and 0.79. CONCLUSIONS: This contrast-enhanced CT-based model can be helpful to early identify poor responder's hepatocellular carcinoma patients and personalize treatments.
OBJECTIVE: The objective of this study is to find a contrast-enhanced CT-radiomic signature to predict clinical incomplete response in patients affected by hepatocellular carcinoma who underwent locoregional treatments. PATIENTS AND METHODS: 190 patients affected by hepatocellular carcinoma treated using focal therapies (radiofrequency or microwave ablation) from September 2018 to October 2020 were retrospectively enrolled. Treatment response was evaluated on a per-target-nodule basis on the 6-months follow-up contrast-enhanced CT or MR imaging using the mRECIST criteria. Radiomics analysis was performed using an in-house developed open-source R library. Wilcoxon-Mann-Whitney test was applied for univariate analysis; features with a p-value lower than 0.05 were selected. Pearson correlation was applied to discard highly correlated features (cut-off=0.9). The remaining features were included in a logistic regression model and receiver operating characteristic curves; sensitivity, specificity, positive and negative predictive value were also computed. The model was validated performing 2000 bootstrap resampling. RESULTS: 56 treated lesions from 42 patients were selected. Treatment responses were: complete response for 26 lesions (46.4%), 18 partial responses (32.1%), 10 stable diseases (17.9%), 2 progression diseases (3.6%). Area-Under-Curve value was 0.667 (95% CI: 0.527-0.806); accuracy, sensitivity, specificity, positive and negative predictive values were respectively 0.66, 0.85, 0.50, 0.59 and 0.79. CONCLUSIONS: This contrast-enhanced CT-based model can be helpful to early identify poor responder's hepatocellular carcinoma patients and personalize treatments.