CT-based radiomics model to distinguish necrotic hepatocellular carcinoma from pyogenic liver abscess

Aim To investigate the feasibility of a computed tomography (CT)-based radiomics model in distinguishing necrotic hepatocellular carcinoma (nHCC) from pyogenic liver abscess (PLA). Material and methods One hundred-four enrolled patients with nHCC (n=56) and PLA (n=48) were divided randomly into a training cohort (n=62) and validation cohort (n=42). ROI (region of interest) of the wall (ROI-wall) and ROI of the necrotic cavity (ROI-necrotic cavity) of the lesion were delineated from each arterial phase (AP) and portal venous phase (PP) image. The least absolute shrinkage and the selection operator logistic regression method was used to select radiomics features, and radiomics scores (R-scores) were calculated. Four radiomics models, including R-score (ROI-wall) in the AP, R-score (ROI-necrotic cavity) in the AP, R-score (ROI-wall) in the PP and R-score (ROI-necrotic cavity) in the PP, were constructed and evaluated by area under the curve (AUC) of receiver operating characteristic curve. Results The AUCs of R-score (ROI-wall) in the AP, R-score (ROI-necrotic cavity) in the AP, R-score (ROI-wall) in the PP, and R-score (ROI-necrotic cavity) in the PP were 0.935 and 0.917, 0.906 and 0.824, 0.985 and 0.928, 0.899 and 0.850, in the training and validation cohorts, respectively. In the training cohort, the AUC of R-score (ROI-wall) in the PP was higher than that of R-score (ROI-wall) in the AP (p=0.024) or R-score (ROI-necrotic cavity) in the AP (p=0.046) or R-score (ROI-necrotic cavity) in the PP (p=0.044). Conclusion CT-based radiomics models can be used to distinguish nHCC from PLA. To investigate the feasibility of a computed tomography (CT)-based radiomics model in distinguishing necrotic hepatocellular carcinoma (nHCC) from pyogenic liver abscess (PLA). One hundred-four enrolled patients with nHCC (n=56) and PLA (n=48) were divided randomly into a training cohort (n=62) and validation cohort (n=42). ROI (region of interest) of the wall (ROI-wall) and ROI of the necrotic cavity (ROI-necrotic cavity) of the lesion were delineated from each arterial phase (AP) and portal venous phase (PP) image. The least absolute shrinkage and the selection operator logistic regression method was used to select radiomics features, and radiomics scores (R-scores) were calculated. Four radiomics models, including R-score (ROI-wall) in the AP, R-score (ROI-necrotic cavity) in the AP, R-score (ROI-wall) in the PP and R-score (ROI-necrotic cavity) in the PP, were constructed and evaluated by area under the curve (AUC) of receiver operating characteristic curve. The AUCs of R-score (ROI-wall) in the AP, R-score (ROI-necrotic cavity) in the AP, R-score (ROI-wall) in the PP, and R-score (ROI-necrotic cavity) in the PP were 0.935 and 0.917, 0.906 and 0.824, 0.985 and 0.928, 0.899 and 0.850, in the training and validation cohorts, respectively. In the training cohort, the AUC of R-score (ROI-wall) in the PP was higher than that of R-score (ROI-wall) in the AP (p=0.024) or R-score (ROI-necrotic cavity) in the AP (p=0.046) or R-score (ROI-necrotic cavity) in the PP (p=0.044). CT-based radiomics models can be used to distinguish nHCC from PLA.