Objectives: To investigate the performance of texture analysis in characterizing P53 mutations of hepatocellular carcinomas (HCCs) based on computed tomography (CT).
Methods: A total of 63 HCC patients underwent CT scans and were tested for P53 mutations. Patients were divided into two groups of P53(-) and P53(+) according to the P53 scores. First- and second-order texture features were computed from the CT images and compared between groups using independent Student's t-test. A Spearman's correlation coefficient was used for correlations to assess the relationship between the different P53 sores and CT data. The performance of texture features in differentiating the P53 mutations of HCC was assessed using receiver operating characteristic analysis.
Results: The mean values of angular second moment (ASM; mean = 0.001) and contrast (mean = 194.727) for P53(-) were higher than those of P53(+). Meanwhile the mean values of correlation (mean = 0.735), sum variance (mean = 1,111.052), inverse difference moment (IDM; mean = 0.090), and entropy (mean = 3.016) for P53(-) were lower than those of P53(+). Significant correlations were found between P53 scores and ASM (r = - 0.439), contrast (r = - 0.263), correlation (r = 0.551), sum of squares (r = 0.282), sum variance (r = 0.417), IDM (r = 0.308), and entropy (r = 0.569). Five texture parameters (ASM, contrast, correlation, IDM, and entropy) were predictive of P53 mutation status, with areas under the curve ranging from 0.621 to 0.792.
Conclusions: There was a direct relationship between P53 mutations and gray-level co-occurrence matrix, but not with histograms for HCC patients. Correlation and entropy seemed to be the most promising in differentiating P53 (-) from P53(+).
Georg Thieme Verlag KG Stuttgart ยท New York.