Title:Gd-EOB-DTPA-enhanced MRI Image Characteristics and Radiomics
Characteristics Combined with Machine Learning for Assessment of Functional
Liver Reserve
Volume: 20
Author(s): Xin-Yu Zhu, Yu-Rou Zhang and Li Guo*
Affiliation:
- Department of Radiology, The Second Affiliated Hospital of Kunming Medical University, No. 374 Yunnan-Myanmar Avenue, Kunming, Yunnan
Province, China
Keywords:
Gd-EOB-DTPA, Radiomics, Functional liver reserve, Cirrhosis, MRI, Machine learning methods.
Abstract:
Objective:
To investigate the feasibility of image characteristics and radiomics combined with machine learning based on Gd-EOB-DTPA-enhanced MRI for
functional liver reserve assessment in cirrhotic patients.
Materials and Methods
123 patients with cirrhosis were retrospectively analyzed; all our patients underwent pre-contrast MRI, triphasic (arterial phase, venous phase,
equilibrium phase) Gd-EOB-DTPA dynamic enhancement and hepatobiliary phase (20 minutes delayed). The relative enhancement (RE) of the
patient's liver, the liver-spleen signal ratio in the hepatobiliary phase (SI liver/ spleen), the liver-vertical muscle signal ratio in the hepatobiliary
phase (SI liver/ muscle), the bile duct signal intensity contrast ratio (SIR), and the radiomics features were evaluated. The support vector machine
(SVM) was used as the core of machine learning to construct the liver function classification model using image and radiomics characteristics,
respectively.
Results:
The area under the curve was the largest in SIR to identify Child-Pugh group A versus Child-Pugh group B+C in the image characteristics, AUC =
0.740, and Perc. 10% to identify Child-Pugh group A versus Child-Pugh group B+C in the radiomics characteristics, AUC = 0.9337. The efficacy
of the SVM model constructed using radiomics characteristics was better, with an area under the curve of 0.918, a sensitivity of 95.45%, a
specificity of 80.00%, and an accuracy of 89.19%.
Conclusion:
The image and radiomics characteristics based on Gd-EOB-DTPA-enhanced MRI can reflect liver function, and the model constructed based on
radiomics characteristics combined with machine learning methods can better assess functional liver reserve.
