Characterization of the Liver Parenchyma Using Parametric T1 and T2 Magnetic Resonance Relaxometry

Although liver biopsy is the current standard for histological characterization of the liver parenchyma, this invasive procedure has a significant risk of - potentially lethal -bleeding. Moreover, as liver disease may focally or heterogeneously affect the liver, histological findings may be false negative or not be representative. Non-invasive imaging modalities (e.g. ultrasound (US), computed tomography (CT) and magnetic resonance imaging (MRI), on the other hand, provide valuable information with regard to global and regional tissue characterization, and may be of help in guiding/targeting liver biopsy. Moreover, in recent years new developments have opened the door towards a better, more comprehensive appreciation of the status of the liver. One of these is elastography assessing the mechanical, viscoelastic properties of the liver, thus providing valuable information with regard to the liver stiffness (or elasticity), and be used for accurate staging of liver fibrosis (1,2). Another approach is tissue characterization using the relaxation behaviour of liver tissue in a high-field magnetic environment ('relaxometry'). As relaxation times are tissue specific, focal or diffuse pathology may alter (i.e. shorten or prolong) relaxation times. These alterations can be used to depict and to quantify disease severity. In the field of cardiac imaging relaxometry has caused a paradigm shift, but for liver imaging this has received relatively limited interest so far. Best known is T2* relaxometry allowing to diagnose and follow-up patients with haemochromatosis. Recently, a few papers have described the use of T1 and T1-rho relaxometry to diagnosis liver congestion in Fontan patients, and to depict liver fibrosis, respectively (3,4).

As part of a comprehensive cardiac MRI exam, we started performing routinely T1 and T2 relaxometry of the heart in 2014. The technique was validated against a standardized, commercially-available phantom in an inter-national study (5). Practically, T1 and T2 mapping includes measurements in both cardiac short-axis and horizontal long-axis before and after the administration of intravenous commercially available gadolinium chelates. As in cardiac short-axis direction the liver is partially encompassed, we will use these images to measure T1 and T2 relaxation times of the liver. As all data are stored on PACS, it is our aim to re-use these T1 and T2 mapping studies to determine the T1 and T2 relaxation of the liver in normal conditions.

Firstly, we will derive from our cardiac MRI database a representative group of studies labelled as normal (i.e. normal cardiac MRI findings). In the medical patient file the serum biomarkers for liver disease will be checked (see below), as well as findings at echocardiography (exclusion of right heart failure, exclusion of severe tricuspid insufficiency), abdominal ultrasound / computed tomography (exclusion of liver disease, i.e. hemochromatosis, steatosis, hepatic congestion, and liver cirrhosis). Only MRI studies will be included if liver disease is excluded. In a next step, a representative region of interest will be manually drawn (> 100 pixels) in a region not including the liver vessels. These analyses will be performed on the pre- and post-contrast T1 map, and on the T2 map. The goal is to determine normal values in at least 100 subjects, allowing to assess the impact of aging and gender.

Secondly, we will measure liver T1 and T2 values in patients with different forms of right heart failure, and assess the relation between liver T1 and T2 values and findings at echocardiography, serum biomarkers, and right heart catheterization indicative of right heart failure. The latter information will be retrieved from the medical patient file. As in right heart failure, the filling pressure increases in the caval and hepatic veins, hereby causing hepatic congestion. It is the hypothesis that in these circumstances T1 and T2 liver values are increased. Also, in latter stages when liver fibrosis initiates and evolves towards cardiac cirrhotic liver, we hypothesize increased T1 values. The aim is to evaluate whether mean T1 and/or T2 of the liver parenchyma obtained at MRI can be used as imaging biomarker of right heart decompensation. As concomitant liver disease (see above) may hamper correct interpretation of our findings, pre-existing liver disease needs to be excluded. For this, a similar approach as in the normal population will be used. The target patient population is three-fold. First, patients with pulmonary hypertension with/without evidence of right heart failure, who underwent a right heart catheterization with invasive pressure measurements (if not available, pulmonary artery pressure estimations at transthoracic echocardiography will be used), and signs of right heart failure at transthoracic echocardiography can be used. Second, patients with dilated cardiomyopathy, defined as LV ejection fraction < 35% with and without concomitant right ventricular (RV) dysfunction (i.e. RV EF < 35%). Third, patients with constrictive pericarditis or inflammatory pericarditis with constrictive physiology (i.e. increased respiratory-related ventricular coupling/interdependence). This group will be compared to a group of patients with inflammatory pericarditis without constrictive physiology (i.e., preserved ventricular coupling).

Thirdly, to assess the inter- and intra-observed reproducibility of liver T1 / T2 values, measurements will be performed by two readers independently performing the measurements. One reader will repeat the measurements respecting one week interval between the analyses. This group involves 10 randomly selected studies in the normal group and 10 studies in the patient group. Analysis include assessment of the intraclass correlation coefficient (ICC) and coefficient of variation (CoV).

Finally, as step towards a more automated approach, we will evaluate/develop texture analysis. This mathematical approach looks at patterns between pixels not visible by the human eye and results in image 'features' that go beyond the typical mean (or median) and deviations. It has shown to be a robust technique in many applications in the field of medical imaging, and most likely will be useful for liver imaging as well. The large group of derived image features needs to be further analysed using machine learning approaches (with/without other clinical readouts). Machine learning of features coupled to a diagnosis ('target') has the potential to augment traditional risk scores with novel imaging biomarkers. Regularization approaches based on e.g. support vector machines, random forests or convolutional neural networks (CNNs) will be implemented.