Liver MRI: Radiomics Features for Prediction of Early HCC Recurrence
Peritumoral radiomics features from gadoxetic acid-enhanced MR imaging have the potential to predict early recurrence in patients with HCC after resection.
Hepatic resection remains the first-line curative treatment for hepatocellular carcinoma (HCC). However, the high recurrence rate of about 70 percent is a major concern. Several risk factors have been identified as indicators for early recurrence, such as poor tumor differentiation or microvascular invasion.
Sunyoung Lee et al. (Journal of Hepatology 2017) found that a combination of two or more significant MRI findings (arterial peritumoral enhancement, non-smooth tumor margin, and peritumoral hypointensity on hepatobiliary phase imaging) is associated with early recurrence after surgery of single HCC
Recently, a literature-based study by Hang-Tong Hu et al. (Abdom Imaging 2018) revealed a significant association between microvascular invasion and peritumoral enhancement, and peritumoral hypointensity on hepatobiliary phase (HBP) imaging.
Radiomics features may be able to gain more information about tumor characteristics compared to visual morphologic analysis. Therefore, Zhen Zhang from Sichuan University, Chengdu, China, and her colleagues aimed to identify potential radiomics features from peritumoral tissue on gadoxetic-enhanced MRI that could predict early HCC recurrence.
In this prospective study, 39 consecutive patients with surgically proven HCC underwent preoperative gadoxetic-enhanced MRI. Follow-up consisted of monthly blood tests and ultrasound or contrast-enhanced CT at one, four, and seven months after surgery. Early recurrence was defined as intra- or extrahepatic recurrence within one year after surgery. Clinical-pathological variables such as age, gender, AFP level (alpha-fetoprotein), and histologic tumor grade were collected.
Qualitative analysis included peritumoral enhancement, peritumoral hypointensity on HBP images, non-smooth tumor margins, and tumor capsule. Quantitative analysis evaluated tumor size, signal intensity ratios, ADC values (diffusion weighted imaging), HBP signal intensity, and radiomics features.
Tumor segmentation was performed manually in images from multiple phases (T2-weighted, non-enhanced T1-weighted, arterial phase, portal venous phase and HBP). The peritumoral region was defined as the area 1 cm away from tumor margin. A total of 385 radiomics features were captured for each sequence. Statistical analysis was performed to find significant differences in radiomics features between recurrence and non-recurrence group.
In the 39 enrolled patients, 19 (48.7%) recurrences were diagnosed based on clinical examination and imaging evidence. 14 recurrences were intrahepatic, 5 were extrahepatic. Median follow-up time was 8.5 months (7-14 months).
Four radiomics features were significantly related to recurrence – two of them derived from T2-weighted imaging (Histogram Min Intensity and Inverse Difference Moment) and two from arterial phase images (Haralick Correlation and Inverse Difference Moment).
A radiomics signature with these features demonstrated good recurrence prediction with an 85% sensitivity, 92% specificity and 0.926 area under the ROC curve (AUC). Multivariate logistic regression analysis identified independent predictive factors for early recurrence:
- Radiomics signature
- Microvascular invasion
- Histologic grade
The combined model comprising these predictive factors showed an improved predictive performance (AUC=0.975) compared to the radiomics signature.
Peritumoral radiomics features had the potential of predicting early recurrence in patients with HCC after resection. Incorporating the clinical-pathological risk factors, the performance of the combined model improved. “Further studies with larger sample size and multicentric collection are still needed”, concluded Zhang.
Presentation Title: Peritumoral radiomics features from gadoxetic acid-enhanced MR imaging predicting early recurrence of hepatocellular carcinoma
Speaker: Zhen Zhang, Sichuan University, Chengdu, China
Date: Friday, March 1st
Session code: B-1044