Texture Analysis Of Magnetic Resonance Images To Predict The Histological Grade And Ki67 Expression In Hepatocellular Carcinoma

Aims: This study aimed to determine whether texture analysis of preoperative magnetic resonance(MR)images could predict the histological grade and expression of Ki67,and help to provide imaging evidence for making preoperative treatment planning.Methods: 108 patients with hepatocellular carcinoma confirmed by histopathology were retrospectively analyzed in our hospital from September 2014 to December 2017,and 83 eligible patients were included follow the inclusion criteria and exclusion criteria.Morphological manifestations of HCC were independently assessed by two radiologists,including arterial phase hyperenhancement,tumor shape,restricted diffusion weight imaging or not,etc.The histological grading of HCC was classified into grade I-IV according to the Edmondson-Steiner grading,in which grade I and II were classified as "low grade group" and grade III and IV were classified as "high grade group".The study population was divided into "low Ki67 group"(Ki67?10%)and "high Ki67 group"(Ki67>10%)according to the expression level of Ki67.All preoperative magnetic resonance data were imported into Ma Zda software to manually delineate the area of interest,and extracted the texture features of the lesions by the images of hepatobiliary phases,T2 WI,arterial phases and portal phases.The Fisher coefficient,POE+ACC,and mutual information(MI)were used to screen out 30 optimal texture parameters.Spearman correlation analysis and ROC analysis were conducted on the obtained parameters by using statistical methods.Results: In all 83 patients with HCC were included(72 males and 11 females;Median age,52y;Range,23-85y).There were 54 cases in the low grade group and 29 cases in the high grade group;25 cases in the low Ki67 group and 58 cases in high Ki67 group.Age between the low Ki67 group and high Ki67 group was statistically significant(P=0.032;P<0.001).For the evaluation of imaging characteristics,the differences of tumor size,irregular shape,hemorrhage and arterial peritumoral enhancement in the two groups were statistically significant(P<0.05),and the consistency between observers was 0.631-0.888.However,the arterial phase hyperenhancement and peritumoral hypointensity on HBP presented statistical differences between the low and high grade groups,and the consistency between observers was 0.708 and 0.482,respectively.The parameters of texture features were maximum extracted based on arterial phase images and Ki67 expression level which the difference are statistically significant.Spearman correlation coefficient was-0.423~0.421(P<0.05);the parameters of texture features were minmum extracted based on portal venous phase images and Ki67 expression level which the difference are statistically significant.Spearman correlation coefficient was 0.240~0.282(P<0.05).The maximum area under the curve of the high and low grade groups was based on the texture feature Perc.10% of the portal venous phase images,and the area under the curve was 0.755.The difference between high Ki67 group and low Ki67 group was based on the texture feature Teta2 of the arterial phase images,and the area under the curve was 0.765.Conclusions: Texture analyses of preoperative magnetic resonance images were helpful for predicting the histological grade and Ki67 expression.It may provide less-invasive achievement of critical histopathology markers for HCC.And it may be also helpful to evaluate the prognosis and response after treatment.