Abstract 1294: Preoperative prediction of microvascular invasion in HCC using radiomics on multisequence gadoxetic acid-enhanced MR images

Abstract

Abstract Micro-vascular invasion (MVI) is underscored as a judicious risk factor for early-stage recurrence and poor prognosis of hepatocellular carcinoma (HCC). Preoperative knowledge of MVI would assist in therapy decision making beforehand, thus prolong the survival of HCC patients. However, current diagnosis of MVI could only be obtained by pathological confirmation after the surgery or through punctures, which hinders timely therapeutic regime plan or causes needless suffering for the patients. In this study, we aimed to perform a preoperative prediction on MVI using multi-sequence Radiomics on gadoxetic acid-enhanced MR images, along with multi-variable clinical factor analysis. A cohort of 208 patients diagnosed with HCC was enrolled with institutional review board approval. Radiomic features including shape and size, intensity, textural and wavelet features were extracted on segmented region of interest. The least absolute regression modeling was used to select the most effective radiomic features which had potential for MVI identification. We finally utilized logistic regression modelling to generate the single radiomics signature for each sequence with Akaike information criteria as the stopping rule. Gadoxetic acid-enhanced hepatobiliary phase (HBP) T1-weighted image and HBP T1 map turned out to be the best-performed sequences with AUC over 0.7 on both training and validation cohorts. The fusion radiomics signature incorporating effective features from these two sequences achieved the optimal prediction with AUC of 0.895 on the training cohort and 0.837 on the validation cohort. Adding the pertinent clinical and radiological factors: AFP, irregular tumor border and arterial peritumoral enhancement, the final combined model could successfully predict the MVI with AUC of 0.943 on the training cohort and 0.861 on the validation cohort. Our study revealed the archetypal radiomic features related MVI and highlighted the fusion radiomics signature as a powerful imaging marker for MVI preoperative prediction. Radiomics combined with traditional clinic-radiological information would no doubt improve the clinical decision making in HCC therapy. Citation Format: Jingwei Wei, Dongsheng Gu, Di Dong, Shuaitong Zhang, Yushen Jin, Jie Tian. Preoperative prediction of microvascular invasion in HCC using radiomics on multisequence gadoxetic acid-enhanced MR images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 1294.