Magnetic resonance texture analysis for the identification of cytokeratin 19-positive hepatocellular carcinoma.

PURPOSE

To investigate potential findings associated with cytokeratin 19 (CK19)-positive HCC, with special emphasis on MR texture analysis.

MATERIALS AND METHODS

Forty-eight patients with CK19-negative HCC and 38 patients with CK19-positive were retrospectively evaluated by texture analysis based on conventional MRI. Clinicalpathological characteristics, conventional MR imaging findings, and the MR texture analysis contained of 2415 texture features in the seven conventional sequences were compared. Significant features for differentiating were identified by univariate and multivariate analyses. Receiver operating characteristic analyses of the significant findings were performed and compared to evaluate their diagnostic performance.

RESULTS

There was no significant difference between the top1 texture feature (three-dimensional standard deviation separation of intensity on T2-weighted original images, abbreviated as: StdSeparation 3D) and the combined top1-6 feature in identifying CK19-positive HCC(P = 0.660). Univariate and multivariate analyses indicated that serum alpha-fetoprotein (AFP) level ≥400 ng/mL(P = 0.013), arterial rim enhancement(P = 0.005), and StdSeparation 3D texture character(P = 0.002) were independent variables associated with CK19-positive HCCs. The combination of the three indices showed a better performance than AFP level(P = 0.0028), arterial rim enhancement(P < 0.0001), and their combination(P = 0.0098); while no significantly better than the StdSeparation 3D texture character alone(P = 0.0788). An acceptable discrimination(AUC = 0.765) with both sensitivity and specificity greater than 75% was achieved for StdSeparation 3D texture character.

CONCLUSION

Serum AFP level ≥400 ng/mL, arterial rim enhancement, and the StdSeparation 3D texture character were independently associated with CK19-positive HCC. The StdSeparation 3D texture character may be a reliable imaging biomarker which can improve the diagnostic performance.