Prediction of the treatment response with texture analysis in patients with hepatocellular carcinoma who have been treated with transarterial radioembolization

Aim In our study, it was aimed to predict the response of hepatocellular carcinoma (HCC) to transarterial radioembolization (TARE) treatment with the parameters to be obtained by the application of texture analysis (TA) on the computed tomography images obtained before treatment in patients with HCC treated with TARE. Material and Method Patients diagnosed with HCC who were treated with TARE at the Radiology Department of Dokuz Eylül University Research and Application Hospital Interventional Radiology unit between January 2010 and January 2020 were retrospectively examined in the imaging archive and patients meeting the criteria were included in the study. TA measurements were performed using the LIFEx program with the specified method and treatment responses of HCC lesions treated with TARE were recorded. Appropriate statistical tests were performed to investigate the relationship between TA measurement values and treatment responses of lesions, to distinguish groups divided according to treatment responses from each other, and to obtain a combined parameter model that can be used to predict the treatment response type before the procedure. Results Kurtosis, Excess kurtosis, GLCM (Energy, Contrast, Correlation, Entropy-log10, Entropy-log 2, Dissimilarity), GLRLM (LRLGE, LRHGE, GLNU, RLNU), NGLDM (Coarseness, Contrast, Busyness), GLZLM (LZLGE, LZHGE, GLNU, ZLNU) parameters were found to differ significantly between the lesion groups with response (RE) and no response (NR) to TARE treatment (p <0.05). Threshold values that can be used to differentiate RE and NR groups were determined as 0.0002, 4166.906, 50.471, 0.001 and 10.360 for the parameters GLCM_Energy, GLCM_Contrast, GLCM_Dissimilarity, NGLDM_Coarseness, and NGLDM_Contrast in ROC (Receiver Operating Characteristics) analysis, respectively. NGLDM_Coarseness with a sensitivity of 81% and a specificity of 71% was determined as the parameter with the largest area under the curve with a value of 0.859 (95% confidence interval, 0.004–-0.002). With logistic regression analysis, a model with the best performance in distinguishing RE and NR groups, including the parameters GLCM_Correlation, GLCM_Entropy_log10, NGLDM_Coarseness and NGLDM_Busyness, was created (p <0.001). The sensitivity of this model in detecting the RE group was 85.9%, specificity 75%, positive predictive value 88.7% and negative predictive value 70%. Conclusion It has been shown that the response to TARE therapy applied in patients with HCC can be predicted before the procedure, thanks to TA. It is thought that TA parameters obtained with standardized methods in large patient groups may contribute to daily radiology practice.