Noninvasive classification of hepatic fibrosis based on texture parameters from double contrast-enhanced magnetic resonance images

Hepatic Fibrosis, the progressive deposition of collagen in the extracellular matrix of the liver, is a fundamental alteration of liver parenchyma in many chronic liver diseases. It impairs liver function, may lead to cirrhosis and hepatocellular carcinoma, and constitutes an important cause of morbidity, mortality, and health care costs (1–4). Early diagnosis is important to initiate treatment and halt progression. Biopsy is the current gold standard for diagnosis, but it is expensive and invasive—factors that limit utility in longitudinal assessments (5–16). The limitations of biopsy for longitudinal assessment have impeded progress in clinical care and research of diffuse liver diseases associated with hepatic fibrosis. There is a need to develop noninvasive imaging techniques that can safely diagnose fibrosis. Conventional ultrasound, computed tomography, and magnetic resonance imaging (MRI) can assess gross liver morphology, or contour, to confirm cirrhosis in patients with advanced liver disease, but such an assessment is insensitive to the early stages of fibrosis and cirrhosis (6,17–19). An alternative imaging-based approach to evaluate hepatic fibrosis is to assess internal liver architecture, or texture. Texture can be defined as a complex visual pattern within an image that consists of simpler subpatterns with characteristic features (20). These features can be assessed objectively by quantitative texture analysis (TA). Quantitative TA has been used to noninvasively classify the liver in a dichotomous fashion as normal or cirrhotic on unenhanced MR images (21,22). In a study with 13 healthy volunteers and 5 cirrhotic patients, quantitative TA of high spatial resolution T2-weighted images provided 100% sensitivity and specificity for discrimination between normal volunteers and cirrhotic patients (21). In another study of 43 cirrhotic patients and 10 normal volunteers, quantitative TA on unenhanced T2-weighted MR images classified patients as cirrhotic or normal with a misclassification rate of 8% (22). Both studies used a publicly available, free software program (MaZda, v. 3.20, Instytut Elektroniki Politechnika, Lodza, Poland) to compute TA parameters from liver images and to generate multivariate predictive models from the parameters (23–25). Because double contrast-enhanced MRI shows fibrosis with greater clarity than unenhanced imaging (26–30), we hypothesized that quantitative TA of double contrast-enhanced MRI may permit accurate dichotomous classification of fibrosis in clinical patients with a spectrum of fibrosis severity, not just discrimination between healthy volunteers versus patients with cirrhosis. The purpose of this pilot, retrospective study was to demonstrate proof of concept that noninvasive quantitative TA on double contrast-enhanced MR liver images can classify liver fibrosis dichotomously in clinical patients, using histology as the reference standard. We used the same software program used in prior studies (MaZda) to compute TA parameters and applied statistical techniques to develop classification models based on the TA parameters.