Perhaps the most common imaging modality to assess liver fat is conventional US, which is widely available and can be performed quickly, cheaply, and safely [53]. However, conventional US only produces a qualitative estimate of liver fat, relies on a subjective interpretation of features in the US image, and is both operator and machine dependent [54–56]. Thus estimates of liver fat derived from conventional US are semiquantitative and have lower repeatability and reproducibility than MRI or MRS and so are of limited value in assessing hepatic fat to inform drug development decisions. US is more reliable at identifying moderate-to-severe degrees of hepatic steatosis as confirmed by histology than are measurements of serum liver transaminase levels [15]. Advanced US methods are being investigated to provide quantitative estimates of liver fat [57]. However, these techniques are still in development and have only limited availability. While these techniques should provide more accurate estimates of liver fat than conventional US, only limited data are currently available on the accuracy and precision of these techniques. Hence, these investigational techniques currently have limited applicability in drug development.

Unenhanced CT evaluates liver fat by measuring x-ray attenuation; x-rays are attenuated less in fatty liver than in non-fatty liver [53]. While CT lacks the ease and portability of ultrasound, it is more widely available and more rapid than MR scans. Unlike US, x-ray attenuation is not subjective and can be measured precisely [58–60]. However x-ray attenuation is affected by several patient factors in addition to fat, reducing the accuracy and precision of liver fat estimation [58]. There may also be variability of attenuation values between CT scanners, particularly those produced by different manufacturers [61]. Finally, CT uses ionizing radiation, which limits its use in vulnerable groups such as children and pregnant women or in studies where multiple scans are required over time [62]. For these reasons CT has only limited applicability for liver fat quantification in drug development, and the use of CT for this purpose is not justified if MRI is available or unless CT of the liver is being performed anyway for other reasons.

The two noninvasive MR-based techniques used to assess liver fat are ¹H (i.e., proton) MRS and MRI. Although the information that these techniques produces is different, the two approaches are similar; the same physics principles underlie and the same equipment is used for both techniques. The patient is placed in a strong magnetic field (called the “main” magnetic field), and then both water and fat molecules in the patient’s body are excited with radiofrequency (RF) pulses. In MRI, gradients (additional magnetic fields of far smaller strength than the main magnetic field) and Fourier transformation are used to “encode” signals allowing images to be produced.

For MRS, gradients are used to locate (i.e., preferentially excite only) a region of interest, and then the Fourier transform of that signal produces a spectrum (which, as described further below, depicts the MR signal intensity transmitted from hydrogen protons in different chemical environments such as those in water molecules and those in the various moieties of triglyceride molecules). To the patient undergoing an MR examination, the difference between the two approaches may not be obvious, except that MRS sequences often are quieter than MRI sequences. Even though MRS measurements technically are not MRI per se, MRS examinations always require some imaging to allow accurate placement of the volume from which fat is being measured. Hence, the phrase “MRI” is often colloquially used to describe both MRI and MRS, even if the main outcome is an MRS measurement.

Most of the strengths and weakness of MRI apply equally to MRS. The lack of ionizing radiation in MR examinations safely permits repeat scanning and hence longitudinal monitoring. It also allows scanning of vulnerable groups, such as children and, if necessary, pregnant women. Given the increasing prevalence of obesity in children, with its downstream problems related to NAFLD, NASH, and other metabolic syndrome complications, the ability to scan children safely, accurately, and precisely to measure fat over time is a significant strength of both MRI and MRS.

A common contraindication to MR scanning is claustrophobia. MR systems are designed to achieve a highly uniform main magnetic field, which is required for both MRI and MRS. The most common scanner geometry that provides this is a long-cylinder (or long “bore”) geometry. Modern MR systems generally have shorter bores than early scanners, decreasing claustrophobia-induced examination failure. Recent wider-bore MR systems have been designed which further reduce the incidence and severity of claustrophobic reactions and also increase the size of patients who can be scanned, which is of importance in conditions associated with obesity. Open systems have also been designed which further reduce claustrophobic reactions, but these systems operate at lower fields strengths and have poorer uniformity than equivalent smaller-bore systems. As yet the advanced MRI and MRS methods detailed in this chapter have not been validated in open systems. Other contraindications to MR scanning arise from the interaction of the (necessary) magnetic field or RF pulses, with metallic “foreign bodies” in the patient being scanned. Many objects implanted for clinical purposes such pacemakers, metallic cardiac valves, and some surgical clips exclude patients from undergoing an MR scan. Also metal that has entered the body accidentally may be of concern, particularly metal fragments in the orbits since the MRI magnet can exert a pull on these tiny metal fragments potentially leading to serious injuries. For this reason, screening is required beforehand, generally in the form of a questionnaire to ensure that patients can safely be scanned.

Both MRS and MRI exploit differences in the resonance frequency of hydrogen nuclei in water and fat molecules to measure fat signal quantitatively. Typical MR spectra from adipose and liver tissue are shown in Fig. 4.1, displaying the separate frequencies, or chemical shifts, of the various water and fat peaks. The resonant frequency of each hydrogen nucleus is determined by the magnetic field it “experiences.” The frequency position of a particular peak, or chemical shift, is determined by the amount of electron shielding it experiences. Both protons in water molecules experience the same amount of shielding and hence the same magnetic field, so there is only one water spectral peak. Different hydrogen nuclei in fat molecules experience slightly different levels of shielding and hence experience different magnetic fields, resulting in a fat spectrum that has several peaks [63]. Individual fat peaks will have a complex shape due to j-coupling, but a discussion of this affect is beyond the scope of this chapter [47]. While the individual main fat peaks can be partially discerned in pure fat tissue, there is insufficient spectral resolution at clinical field strengths (1.5 T and 3 T) to delineate all of the liver fat peaks in vivo [63].

The resonant frequency of each MR spectral peak (fat, water, and other) could be, but rarely is, given as its actual resonant frequency (near 63 MHz for 1.5 T scanners and near 127 MHz for 3 T scanners). However, it is useful and is common practice to indicate the position of each spectral peak in parts-per-million (ppm), which is independent of the strength of the main magnetic field. By convention, spectral peak frequencies are measured with respect to a reference material (tetramethylsilane (TMS)), which is assigned a frequency of 0 ppm. In vivo, there will be no TMS present, and so the water spectral peak acts as a surrogate reference. Water has a chemical shift (i.e., spectral peak position) of 4.7 ppm at 37 °C. Different types of fat will have slightly different spectra, depending on the amount of saturated, monounsaturated, and polyunsaturated fat they contain. Several studies have shown that MRS and MRI both can be used to determine “type” of fat – i.e., the degree and type of unsaturation of fat [63–65]. However, description of the details of these more advanced methods is beyond the scope of this chapter; we will focus here only on MR measurement of fat content.

MRI and MRS both measure the amount of fat by referencing the MR signal from fat to the MR signal from water. To be of utility, the fat and water signal values are converted to a value which is independent of possible confounding effects present in both MRI and MRS, some of which arise from and can be avoided or corrected for by judicious choice of MR scanner settings. When placed in a magnet field, hydrogen nuclei align with the main magnetic field to create longitudinal magnetization (M _z ). This magnetization is not detectable when a patient is just placed in the main magnetic field, so an “excitation” RF pulse at the resonant frequency of the hydrogen nuclei is applied which transfers energy to (excites) M _z , rotating it into the transverse plane where it begins to precess (rotate), as shown in Fig. 4.2. The magnitude of the magnetization in the x–y plane is called M _xy . When excited into the x–y plane, the precession frequency of the magnetization vector as it rotates at the resonant frequency is given by the Larmour equation:

where ν is the Larmour (resonant, precession) frequency, γ is the gyromagnetic ratio (which is different for each type of nucleus), and B ₀ is the main magnetic field that each atomic nucleus experiences (i.e., after shielding).

The signal radiated by the excited protons will exponentially decay with a (transverse) rate of decay (or relaxation) given by T2, as described in the equation:
where M _xy (0) is the initial transverse magnetization before it starts to decay, the echo time (TE) is the time between the RF pulse and the signal acquisition, and T2 is the time for M _xy to decay to 1/e of its initial value (Fig. 4.3). In MRS, we have to correct for T2 decay, but in the MRI sequences discussed in this chapter, this decay is described by T2*, rather than T2, which accounts for additional sources of relaxation that protons typically experience during imaging. MR images require numerous excitations, often hundreds to thousands, and the amount of M _z (i.e., ready to be tipped again into the x–y place to give more signal) after an excitation is given by the equation:

where TR is the time between acquisitions and T1 is the time required for M _z to recover to (e − 1)/e of its original value (Fig. 4.4).

Water and fat have different values of T1, T2, and T2*. These values will vary from patient to patient and also are dependent on the field strength of the MR scanner [66]. Thus different sequences with different TR and TE values may give different estimates of the fat content. Choosing sequence parameters that minimize T1 and T2 weighting, or acquiring data such that these parameters can be corrected, can avoid or correct for these patient-related differences. This chapter will discuss several of these, as well as other confounding factors, and discuss how to produce an estimate of fat that is independent of field strength and scanner platform.

MRS is widely considered to be the most accurate, noninvasive method to estimate hepatic PDFF [53]. As shown in Fig. 4.1, MRS at 1.5 and 3 T can directly identify the main spectral peaks in fat and water. The amount of signal associated with a spectral peak is related to the area under the peak. Thus, if there were no confounding effects, summing the areas under all the spectral fat peaks and also measuring the area under the water peak would produce an estimate of the ratio of fat-to-water signal, which is related to how much fat is present. However, as stated above, these signals can be confounded by several factors. To produce a repeatable, reproducible estimate of amount of fat, independent of confounding factors, specialized acquisition, and analysis procedures are required. As a first step, MR images of the liver are acquired to allow a volume of the liver to be selected. Normally, images are acquired in more than one plane to assist in selecting an MRS volume of interest, also known as a “voxel.” The MRS voxel, generally a 2 × 2 × 2 cm cube, is manually placed in liver parenchyma, using the localizing MR images, to avoid liver edges, large vessels, and large bile ducts as shown in Fig. 4.5. Shimming, a series of short tuning scans, is then performed to achieve a maximally homogeneous magnetic field across the MRS voxel. After successful shimming, an MRS sequence is performed.

There are two main MRS sequences used to measure liver fat. These are PRESS (point resolved spectroscopy) [67] and STEAM (stimulated echo acquisition mode) [68, 69]. There are advantages to both these sequences. PRESS provides twice the signal to noise obtainable from STEAM, but STEAM is capable of a shorter echo time (TE) [70]. Both sequences have been used to measure hepatic PDFF, but as discussed later in this section, STEAM is preferable to PRESS for liver fat quantification. When measuring liver fat using MRS, no water or fat saturation should be used [35]. Also, spatial saturation bands should be disabled. Spatial saturation bands remove all signals from a selected volume band and are used in spectroscopy to remove possible contamination from outside the voxel but can also introduce an unequal response across the spectrum within the selected MRS voxel [71]. While previous studies have used free breathing and multiple averages, there is sufficient signal to noise to accurately measure liver in a single breath-hold.

As discussed previously, water and fat have different T1 and T2 values, and these vary between individuals. In MRS, to remove T1-induced variability (mainly of water), TR >3,000 ms is chosen. This TR is sufficient to minimize/avoid the natural variability in T1 between individuals, while still allowing multiple spectra to be acquired in a single breath-hold. This long TR also minimizes the effect of the water and fat having different T1 values.

There are two ways to correct for T2 variability between individuals. The simplest is to collect spectra at as short a TE as possible. This minimizes differences in the amount of decay that occurs due to intersubject variability in T2 values. STEAM has a shorter minimum TE than PRESS and so avoids T2 variability better than PRESS. Mean values of T2 for water and fat, taken from literature, are then used to correct approximately for the T2 decay that does occur. A more advanced solution is to collect single-average spectra at multiple TE values [47, 48]. With a TR of 3,500 ms, spectra at five different TE values can be comfortably collected in a single breath-hold. This allows the areas of fat and water peaks to be calculated by fitting the T2 decay equation:
Here S is the area under the spectral peak curve, and S ₀ is the T2-corrected peak area (i.e., the area under the curve that would be present at TE = 0 ms).

The range of TE values for multi-TE T2 measurement must carefully be chosen, as at longer TE values the decay of fat is no longer purely exponential due to j-coupling. This again suggests the use of the STEAM sequence, which is capable of shorter TEs [72]. A typical liver MRS multi-TE acquisition is shown in Fig. 4.6. However, acquisition of spectra at multiple TE values in a single breath-hold requires a specialized version of MRS sequence. For some scanners, the ability to run these specialized sequences may require a research agreement with the scanner manufacturer.

Virtually all liver MRS is effectively limited to single-voxel acquisition. Thus MRS cannot readily provide information about the spatial distribution of fat in the liver. Liver fat distribution can be heterogeneous, and it is possible that the MRS voxel location is not indicative of the liver as a whole. This may pose a challenge in longitudinal studies, as care must be taken to co-localize the MRS voxel in follow-up scans to a similar location as was selected in the initial study. Otherwise spatial variation in the liver may be mistaken for a change in liver fat over time. There are multi-voxel MRS sequences available that cover large volumes of the liver, providing information on the spatial distribution of fat. However, the large volume reduces the uniformity of magnetic field within the volume, producing poor quality spectra. Also multi-voxel sequences are time consuming and beyond the range of single breath-hold techniques. Thus, they are highly susceptible to artifacts introduced by patient breathing, such as contamination of fat signal from surrounding adipose tissue [35, 66].

Automated, reliable, online analysis methods of spectral analysis are not yet available, so spectral analysis must be performed offline with specialized software by experienced personnel [73]. Spectral tools available from MR scanner manufacturers generally are suitable for viewing spectra, but not optimally suited to estimate hepatic PDFF. However there are a number of advanced tools that incorporate prior knowledge of the structure of the spectrum into fitting algorithms [74, 75]. These tools require initial input from a physicist with advanced knowledge of fat-related MRS to set up so-called “prior knowledge” constraints for analysis of fat spectra. This prior knowledge is generated from a number of factors, including biochemical data and MR scanner parameters. Further, these tools are generally designed for research rather than clinical use, and even when prior knowledge has been set up, an experienced user is still needed to analyze the spectra. Finally on many MR scanners, the raw MR data required for MRS analysis is not automatically archived but instead is stored in temporary files that require a research agreement with the scanner manufacturer to access.