Objective:

This retrospective study aims to evaluate the feasibility of gray scale ultrasound image based radiomics analysis in prediction of Ki-67 positive rate in histopathollogically proved hepatocellular carcinoma (HCC).

Methods:

Grayscale ultrasound images of 133 patients that underwent operation and histopathologically proved HCC lesions were analyzed. Ultrasound gray scale images (GS-US) were segmented to extract the wavelet

texture and morphological features of the tumor in the image. Afterwards

234 features were selected by genetic algor

ithm with minimum-redundancy-maximum-relevance (mRMR)

and were further screened by the sparse representation method. The best feature subsets were used for classification.

Results:

GS-US images of HCC lesions were classified and evaluated by the prediction model of support vector machine (SVM) with leave-one-of-cross validation (LOOCV). The area under the receiver operating characteristic curve (AUC) in the results reached 0.75.

Conclusion:

Imaging (based radiomics approach) analysis of GS-US images correlated to the Ki-67 expression positive rate in HCC lesions

which might be helpful in clinical management and prognosis prediction.