Non-alcoholic fatty liver disease (NAFLD) is the most common liver disease world-wide and prevalence is increasing due to the dual epidemics of obesity and type-two diabetes. Hepatic steatosis is a defining feature of NAFLD but accurate quantification is reliant on liver biopsy with its inherent risk of serious complication. There is a need to provide accurate, non-invasive techniques for the assessment of hepatic steatosis for use in clinical practice and as an endpoint in interventional trials. Magnetic resonance spectroscopy (MRS) has been shown to accurately assess hepatic steatosis¹. However, it is not yet routinely used in clinical practice. The reasons for this include a lack of accessible tools and/or expertise in many centres for the appropriate acquisition and analysis of the data. Existing studies in the literature have been restricted to expert single-centres and have largely employed manual or semi-automated analysis methods to measure liver fat². In this study we aim to quantify and characterise liver fat in NAFLD patients in comparison with histology using a novel automated approach to the analysis of liver MRS employing open-source TARQUIN³ software.

Unselected, sequential patients having a standard-of-care liver biopsy for the diagnosis or staging of NAFLD were scanned at two centres (Queen Elizabeth Hospital Birmingham and Clinical Research Imaging Centre, Edinburgh). Table 1 shows the patient cohort characteristics. Histology was assessed by expert pathologists and diagnosis of simple steatosis (SS) or non-alcoholic steatohepatitis (NASH) was by Brunt criteria⁴.

MRS was performed at 3T as part of a comprehensive quantitative liver MRI protocol, using PRESS (TR 3s, TE 30ms) without water suppression (WREF) and STEAM (TR 3s, TE 20ms) with water-suppression (WS) and without water-suppression in a series of three 15 second breath-holds following automated high-order shimming. For STEAM, 5 measurements were acquired for individual processing while the PRESS was acquired as a single measurement with 5 averages. High-resolution localisers in three planes were used to carefully place the voxel (20x20x20 mm) in the right lobe of the liver avoiding large vessels and ducts.

Spectroscopy data were automatically corrected for frequency and phase offsets and fitted using TARQUIN³. A customised basis-set was used to model the time-domain signal of fatty liver with components as shown in table 2. Residual water was not removed from the WS spectra. In all fits, the water peak was modelled with two components at 4.65 ppm. Quality control was performed by visual inspection to exclude spectra with unresolved fat peaks, severe baseline artefacts or bad fits. Simple fat fraction (FF) was calculated without T2 correction using STEAM and PRESS WREF fits. Mean chain length (CL) and number of double-bonds per chain (nDB) were estimated from the STEAM WS fits using a method similar to Hamilton et al². Spearman’s correlation analysis was performed to investigate the relationship between i) PRESS and STEAM, ii) different fat spectral components and iii) FF, CL, nDB and histological markers of disease severity.

Good quality spectra were acquired at two centres in a short scan time (< 2 min including shimming) allowing accurate FF measurements in all but 1 case and lipid profile analysis in all but 3 cases. Figure 1 shows an example of a complete dataset for a patient with grade 2 steatosis showing good quality spectra with separation of fat peaks and accurate fits as demonstrated by the small residuals. Close agreement was found between FF calculated from PRESS and STEAM acquisitions, with PRESS giving a slight over-estimation compared with STEAM (Figure 2) as expected due to T2 effects⁵. Highly significant correlations were found between FF and histological grade of steatosis (r = 0.81, 0.77 for STEAM, PRESS respectively, both P<0.001). Highly significant negative correlations were found between FF and CL (r = -0.57, P<0.001) and between FF and nDB (r = -0.44, P=0.002). In addition, the lipid profile showed negative correlation between nDB and hepatocyte ballooning, independent of FF (r= -0.31, P=0.03). Figure 3 illustrates this potential dependence of lipid profile on disease severity by comparing normalised STEAM WS spectra for example cases of NASH and SS with similar FF of 34% and 42% respectively.We demonstrate that MRS is a powerful non-invasive tool to quantify hepatic steatosis and that automated analysis using TARQUIN is possible, providing strong correlations with histology. Assessment of fatty acid saturation and chain length compared with histological markers of disease severity has not been previously reported using MRS in human liver. This study paves the way for further investigation into the relationship between fatty acid composition and clinically relevant markers of disease severity in NAFLD.