Non-Gaussian diffusion MRI models for preoperative assessment of microvascular invasion in hepatocellular carcinoma

This study aims to assess the potential value of the non-Gaussian diffusion MRI models, including intravoxel incoherent motion (IVIM), stretched exponential model (SEM), diffusion kurtosis imaging (DKI), fractional-order calculus (FROC), and continuous-time random walk (CTRW) models, for preoperative prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC).

A total of 61 consecutive patients were enrolled in this study. Various diffusion parameters were calculated from six diffusion models: conventional diffusion-weighted imaging (DWI) and five non-Gaussian diffusion MRI models (IVIM, SEM, DKI, FROC, and CTRW). The clinical characteristics and various diffusion parameters were analyzed using chi-square tests, independent sample t test or Mann-Whitney U test between subgroup. Logistic regression was applied to explore the risk factors from clinical variables. Diagnostic accuracy was evaluated through the area under the receiver operating characteristic (ROC) curves (AUCs), the Delong test was performed to evaluate the AUCs.

Patients with MVI-positive HCC demonstrated significantly higher DKI_K (P < 0.001), FROC_β (P = 0.047 ) values, and remarkably lower CTRW_α values (P = 0.002 ). Further, multivariable logistic regression analysis indicated that DKI_K (odds ratio (OR) = 1.004, P = 0.044) and FROC_β (OR = 1.005, P = 0.027) values were significantly associated with MVI-positive HCC, with higher AUC values compared with ADC values (0.746 vs. 0.649 vs. 0.596, respectively). Additionally, the alpha-fetoprotein (AFP) levels (OR = 1.002, P = 0.009) and maximum tumor diameter (MTD) (OR = 1.001, P = 0.004) were also included as risk factors for MVI-positive HCC. Combining DKI_K and FROC_β values with the AFP levels and MTD improved the AUC value to 0.878.

DKI_K and FROC_β values had the potential to serve as independent predictors of MVI in HCC, and their combination with MTD and AFP levels further enhanced the performance.