Whole-lesion iodine map histogram analysis combined with focal ROI iodine concentrations for preoperative prediction of cytokeratin 19-positive hepatocellular carcinoma.

PURPOSE

To assess the value of whole-lesion iodine map (IM) histogram analysis and focal region-of-interest (ROI) iodine concentrations (ICs) in preoperative prediction of cytokeratin 19 (CK19)-positive hepatocellular carcinoma (HCC).

METHODS

We retrospectively analyzed 82 patients with HCC who underwent spectral computed tomography (CT). Based on their CK19 status, the patients were categorized into positive and negative groups. The whole-lesion IM histogram parameters (mean, standard deviation [SD], variance, coefficient of variation [CV], skewness, kurtosis, entropy, and 1st, 10th, 25th, 50th, 75th, 90th, and 99th percentiles) were generated by manually delineating the entire tumor region for each HCC. In other methods, IC was measured from IM by placing ROIs at representative tumor levels and normalized to aortic IC. The t-test, Mann-Whitney U, and χ test were used for group comparisons. p-values for histogram parameters were adjusted using the false discovery rate (FDR), and Pearson correlation was used to assess multicollinearity among significant parameters. Receiver operating characteristic (ROC) curve analysis was used to assess diagnostic performance, with AUC comparisons performed using DeLong's test. Internal validation was achieved using a 5-fold cross-validation.

RESULTS

The IC and normalized IC (NIC) of the focal ROI and the SD, variance, CV, entropy, 1st, 10th, and 25th percentiles of the whole-lesion IM histogram displayed significant differences between the CK19-positive and CK19-negative groups (P < 0.001 to P = 0.009). Based on the correlation analysis, only SD, entropy, and the 1st percentile were retained for further analysis. ROC curve analysis demonstrated comparable diagnostic performance of histogram variables and focal ROI ICs (AUC = 0.779, 0.793, respectively). Combining parameters from both focal ROI ICs and histogram analysis provided optimal diagnostic performance for predicting CK19 status (AUC = 0.910). Cross-validation of this combined approach yielded a mean AUC of 0.892, accuracy of 78.1%, sensitivity of 78.6%, specificity of 77.5%, and Cohen's κ of 0.561.

CONCLUSION

Whole-lesion IM histogram parameters and focal ROI ICs present comparable clinical value for preoperatively predicting CK19 expression in HCC, with the combined parameters exhibiting the best performance.