Liver MR relaxometry at 3T – segmental normal T1 and T2* values in patients without focal or diffuse liver disease and in patients with increased liver fat and elevated liver stiffness

Magnetic resonance (MR) T1 and T2* mapping allows quantification of liver relaxation times for non-invasive characterization of diffuse liver disease. We hypothesized that liver relaxation times are not only influenced by liver fibrosis, inflammation and fat, but also by air in liver segments adjacent to the lung – especially in MR imaging at 3T. A total of 161 study participants were recruited, while 6 patients had to be excluded due to claustrophobia or technically uninterpretable MR elastography. Resulting study population consisted of 12 healthy volunteers and 143 patients who prospectively underwent multiparametric MR imaging at 3T. Of those 143 patients, 79 had normal liver stiffness in MR elastography (shear modulus <2.8 kPa, indicating absence of fibrosis) and normal proton density fat fraction (PDFF < 10%, indicating absence of steatosis), defined as reference population. T1 relaxation times in these patients were significantly shorter in liver segments adjacent to the lung than in those not adjacent to the lung (p < 0.001, mean of differences 33 ms). In liver segments not adjacent to the lung, T1 allowed to differentiate significantly between the reference population and patients with steatosis and/or fibrosis (p ≤ 0.011), while there was no significant difference of T1 between the reference population and healthy volunteers. In conclusion, we propose to measure T1 relaxation times in liver segments not adjacent to the lung. Otherwise, we recommend taking into account slightly shorter T1 values in liver segments adjacent to the lung.

However, as known from cardiac applications, reference values should be established for different manufacturers, technical parameters and field strengths before using T 1 mapping in clinical routine 15,16 . Further, it is known from cardiac imaging that T 1 values might be influenced if measured in the ventricle close to the lung, why usually measurements in the septum are preferred 17 . Therefore, specific anatomical conditions of the liver with adjacent air-containing lungs and potential internal confounders, such as hepatic fat and iron composition, should be considered in hepatic T 1 mapping 2 .
We hypothesized that T 1 relaxation time is significantly shorter in liver segments adjacent to the lung than in liver segments not adjacent to the lung due to air-induced susceptibility effects at 3T. Table 1. Patients without steatosis but with increased liver stiffness (shear modulus ≥ 2.8 kPa) showed a male predominance (78% males vs. 43% males in patients without fibrosis), a higher prevalence of diabetes (35% vs. 3%, p < 0.001), elevated GGT (110 ± 143 vs. 30 ± 33, p < 0.001), and prolonged extrinsic coagulation times (Quick 81 ± 20% vs. 99 ± 3%, p < 0.001). There was a tendency for more tobacco smokers in the group with increased liver stiffness, without statistical significance (p = 0.091). None of the patients was taking empagliflozin and only one patient was taking ezetimibe, both known to reduce liver fat as measured with PDFF 18,19 . Image quality. Out of 143 patients and 12 volunteers, resulting in 155 participants, there was at least one segment of good diagnostic image quality in 152/155 participants (98%) for T 1 and in 149/155 participants (96%) for T 2 * (Table 2). In the right liver, segments V-VIII were measurable in at least 90% of the patients for T 1 . For T 2 *, the left liver segments were just measurable in 21-40%, while on the right side, the cranial right segments (VII and VIII) were measurable in 61-63% of patients and the right caudal segments (V and VI) were measurable in 86-88% of patients. Table 3 and demonstrated in Fig. 1, T 1 values were significantly lower in the cranial slices than in the caudal slices (mean of the differences 33 ms, p < 0.001). This did  Table 1. Patient characteristics of the MR elastography study population (patients n = 143, volunteers n = 12, total n = 155). Values represent the mean ± SD or n. P-values were calculated using the Mann-Whitney U or Fisher's exact test, as appropriate. Comparisons between the two patient groups with steatosis and the reference group in the first column are indicated with *p < 0.05; **p < 0.001. MR = magnetic resonance; PDFF = proton density fat fraction; BMI = body mass index; ASAT = aspartate aminotransferase; ALAT = alanine aminotransferase; GGT = gamma-glutamyltransferase; APRI = aspartate aminotransferase-to-platelet ratio index.

Location-based mapping results. As shown in
not count for the comparison between cranial segment II and caudal segment III, where no significant difference between both segments was detected (6 ms, p = 0.365). Values between segments were very comparable on each level, except for segment II, which showed values more comparable with the caudal level segments ( Table 3). The same tendencies, but less pronounced, were observed for T 2 *. Normal T 1 values in reference patients were as follows: T 1 = 780 ± 83 ms (segments not adjacent to the lung) and 746 ± 81 ms (segments adjacent to the lung) ( Table 4). There were no significant differences between the younger healthy volunteers and the reference population (p = 0.358 for T 1 and p = 0.521 for T 2 *).
In positive patients with steatosis as well as in patients with increased liver stiffness, T 1 relaxation time was significantly longer than in the reference population (p = 0.019 and p = 0.011 and p < 0.001 for steatosis only, increased stiffness ≥ 2.8 kPa only and increased stiffness ≥ 3.5 kPa only, respectively). These results are illustrated in Fig. 2.

Multivariate analysis.
In multivariate analysis of patients and volunteers without focal or diffuse liver disease, T 2 * time was a significant confounder of T 1 time (p < 10 −15 ), while PDFF, age and sex (p = 0.249, 0.722, 0.687, respectively) were not (Fig. 3).

Discussion
This study shows that T 1 relaxation time is significantly shorter in liver segments adjacent to the lung than in liver segments not adjacent to the lung. We calculated a mean T 1 value of 780 ± 83 ms at 3T in liver segments not adjacent to the lung (segments II, III, IVb, V, VI), while T 1 was around 30 ms shorter in liver segments adjacent to the lung (segments IVa, VII, VIII) in a patient population, without focal (based on CT) or diffuse (based on PDFF and MR elastography) liver disease.  Table 2. Mapping quality in different liver segments in all study participants (patients n = 143, volunteers n = 12, total n = 155). Values represent the numbers of participant with liver segments with good image quality on T 1 and T 2 * maps, respectively. Relative portion compared to total study population (n = 155) is shown with the % shown in brackets. The first column indicates the number of participants with at least one segment with good image quality, followed by the number of participants with good image quality maps in Segment VI/VII, V/VIII, Iva/B and II/III, respectively. In the upper part of the Table results for T 1 , in the lower part results for T 2 * are described. P-values were calculated using Fisher's exact test to compare the number of segments with good image quality at every localization to the number of at least one segment with good quality per patient.  Table 3. Reference values in different liver segments in patients with normal liver stiffness without steatosis (n = 79). Values represent the mean ± SD or n. P-values were calculated using a paired Mann-Whitney U test to compare upper and lower levels and each liver segment with the mean value of all liver segments of the same level: *p < 0.05; **p < 0.001.
www.nature.com/scientificreports www.nature.com/scientificreports/ This difference may be explained by different reasons. One possibility are susceptibility effects from adjacent air in the lungs. Susceptibility differences between liver and lungs cause off-resonance, which may lead to T 1 underestimation. Similar effects are known from myocardial T 1 mapping 20,21 . In addition, there is a T 2 -dependency of the MOLLI based T 1 mapping sequence due to its bSSFP design, therefore influencing the measured T 1 value. Another explanation might be partial volume effects at the liver dome since the slice thickness is 10 mm (Fig. 4), which is also known from neuroimaging 22 . However, the regions of interests (ROI) were drawn carefully to exclude liver vessels and outer 10 mm of the liver border, to prevent this possible bias as much as possible.
While differences between cranial and caudal segments in the right liver lobe and segment IVa vs. IVb were comparable, the T 1 relaxation time difference between segment II and III was smaller. A potential explanation for this different behaviour might be that segment II is in close anatomical neighborhood to the heart and not to the lungs as the cranial segments VII, VIII and IVb are. We therefore propose to measure T 1 relaxation time in liver segments V and VI with excellent inter-reader-reliability, when a representative T 1 quantification of the liver is intended. These segments are normally targeted for liver biopsy and may be regarded as representative for the whole liver. However, for a more detailed segmental analysis of T 1 relaxation times in the liver shorter normal T 1 values in liver segments adjacent to the lung should be taken into account. T 2 * seems to be more prone to respiratory and pulsation artefacts from the beating heart, which might explain lower image quality in segments of the left liver lobe (adjacent to the heart) and in cranial segments, while image quality in caudal left segment V and VI showed the best image quality. Electrocardiography gating might improve image quality, which was not tested in this study.   Table 4. T 1 values of lung-adjacent and non-lung-adjacent liver segments in all study participants (patients n = 143, volunteers n = 12, total n = 155). Values represent the mean ± SD or n. P-values were calculated using the Mann-Whitney U test, *compared to patients with normal liver stiffness without steatosis, **compared to patients with normal liver stiffness with steatosis. www.nature.com/scientificreports www.nature.com/scientificreports/ volunteers using an inversion recovery method with refocusing pulses 25 , and 745 ± 65 ms in 8 healthy volunteers using MR fingerprinting 26 . Slight differences might be explained by different techniques of assessment and small sample sizes used in the cited studies. This underlines the recommendation, that normative values should be established for the particular site and set-up for different manufacturers, technical parameters and field strengths before using T 1 mapping in clinical routine 15,16 . In accordance with other published studies, increased T 1 time correlated with increased liver stiffness in MR elastography as a non-invasive surrogate for liver fibrosis 2,13,14,23 . Yoon et al. published T 1 values in patients with chronic liver disease (863 ± 81 ms) and Child A liver cirrhosis (879 ± 86 ms) 27 in the same range as we present here (849 ± 107 ms for early liver fibrosis and 910 ± 98 ms for significant fibrosis). T 1 values calculated with commercially available software solutions using the MOLLI mapping technique have been shown to correlate with the severity of NAFLD/NASH and fibrosis 28 and may predict clinical outcome in these patients 29 . We demonstrated an increased T 1 time in patients with steatosis, which is in accordance with other published studies 30 . This is a paradox, since fat has a much shorter T 1 time than water. According to recent publications, off-resonance effects in imaging voxels containing a mixture of fat and water signals may explain this finding 31,32 .
In addition to fat and fibrosis content as well as susceptibility-effects in liver segments adjacent to the lung, there are several other potential confounders of T 1 and T 2 *. One known confounder inducing T 1 shortening is liver iron content 2,33 , however patients with iron overload have been excluded in this study. Liver blood distribution and oxygenation levels are might influence T 1 and T 2 * times as well. These effects and possible influences on T 1 and T 2 * need further investigation. In multivariate analysis of reference patients without focal or diffuse liver disease and healthy volunteers, we could not demonstrate a significant age-or sex-dependency of the measured T 1 values. However, there was a collinearity between T 1 and T 2 * values. Further research should be performed to show whether T 1 is T 2 * dependent (e.g. due to fat, iron composition, blood distribution and eventually blood oxygenation influencing T 2 *) or whether it is the other way round and T 2 * is influenced by T 1 due to a too short TR in the used T 2 *-mapping sequence. Also using a multiparametric approach combining T 1 , T 2 *, MRE and PDFF might deliver further insights.
Our study highlights the importance of standardized technical parameters and well-defined normal values when performing T 1 mapping of the liver. For an accurate non-invasive characterization of diffuse liver disease, a combined analysis of different MR parameters such as T 1 mapping, T 2 * mapping, elastography and PDFF should be performed. Further research will show how a multiparametric combination of those MR imaging biomarkers may help differentiating and quantifying diffuse liver disease.
Limitations. Our study has several limitations. First, we focused on a population without known chronic liver disease. Due to the study design and associated ethical considerations, liver biopsy was not possible. Instead, we used MR elastography and PDFF, which have been shown to correlate very well with biopsy-confirmed fibrosis 34 and steatosis grades 35 , as the non-invasive gold standard, and we included negative and positive controls based on these imaging techniques. Another limitation is that we did not obtain full 3D coverage of the liver, as we used commercially available Siemens MOLLI sequences with 2D acquisitions on three transverse slices. Nevertheless, we present a segmental comparison of T 1 mapping in a reference population, as well as in negative and positive controls. Newer 3D mapping sequences or MR fingerprinting may offer even more applications for future use.

Conclusion
When analysing T 1 maps in the liver at 3T, we propose to measure T 1 relaxation times in liver segments not adjacent to the lung. Otherwise, we recommend taking into account slightly shorter T 1 normal values in liver segments adjacent to the lung. www.nature.com/scientificreports www.nature.com/scientificreports/ Methods study population. This prospective cross-sectional study was approved by the institutional review board (Kantonale Ethikkomission Bern, IRB number 282-15) and was conducted in accordance with relevant guidelines and regulations after obtaining written patient informed consent. All participants underwent multiparametric MR imaging at 3T in our institution between 03/2016-06/2017, including T 1 and T 2 * mapping, proton density fat fraction (PDFF) quantification and MR elastography.
A total of 161 study participants were recruited, while 6 patients had to be excluded due an incomplete MR exam due to claustrophobia or technically uninterpretable MR elastography. Resulting study population consisted of twelve healthy volunteers without a history of liver disease (negative controls) and 143 patients (Fig. 5). The included 143 patients were selected based on acquired abdominal computed tomography (CT) scans without focal liver disease (cysts > 2 cm, solid lesion >1, prior liver surgery). Out of the 143 patients, we then defined a reference population (n = 79) without focal (based on CT) or diffuse liver disease (based on MR elastography shear modulus < 2.8 kPa and PDFF < 10%). CT scans in the reference population were performed with following indications: trauma (n = 14), abdominal pain (n = 19), infection (n = 23) and tumor search (n = 23). The remaining patients (n = 64) were assigned to positive control groups with diffuse liver disease (MR elastography shear modulus ≥2.8 kPa and/or PDFF ≥10%).
Clinical information and laboratory test results were recorded for the included patients. Clinical parameters included age, gender, body mass index (BMI), history of diabetes or hypertension, daily drug intake, tobacco use and alcohol consumption. Biological parameters included dyslipidemia, platelet count, quick value, total bilirubin levels, gamma-glutamyltranspeptidase (GGT), aspartateaminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase, albumin, creatinine and hematocrit.  www.nature.com/scientificreports www.nature.com/scientificreports/ with a multiecho gradient echo (GRE) single breath-hold sequence (12 echoes with a TE between 0.93-14.2 ms, TR of 200 ms, FA 18°, FOV 400, 10-mm slice thickness, and scan time of 19 s). T 1 and T 2 * maps were generated on three single slices in the upper, mid and lower liver. PDFF was calculated using the Dixon method with axial T 1 -weighted axial vibe images (TE of 2.45 ms and 3.68 ms, TR of 5.47 ms, FA 9°, 3-mm slice thickness, and scan time of 22 s) to differentiate patients with and without liver steatosis. For MR elastography, a pneumatic driver (Resoundant) was placed on the right upper quadrant transmitting shear waves by continuous acoustic vibrations with a frequency of 60 Hz. The liver shear stiffness in kPa in the right upper liver lobe was determined with a gradient echo-based elastography sequence (WIP package 622 provided by Siemens Healthineers, 3 single-slice acquisitions with 5-mm slice thicknesses) using the 95% confidence map of stiffness. A shear modulus ≥2.8 kPa was considered to represent early liver fibrosis (corresponding to histology fibrosis grade ≥F1, according to the Metavir staging system), while a shear modulus ≥3.5 kPa was defined as significant liver fibrosis (corresponding to histology fibrosis grade ≥F3) 8,36,37 . MR imaging analysis. Prior to any measurements image quality on relaxometry maps was assessed by an experienced radiologist (V.O., 5 years of experience in hepatic imaging). For T 1 and T 2 * mapping, 8 polygonal regions of interest (ROI) were drawn in liver segments II-VIII by the radiologist (V.O.) who was blinded to the patient's clinical history. In liver segments without excellent image quality (e.g. due to motion artefacts) or in segments that were not captured on any of the three acquired slices, no ROI was drawn at this location, and thus no value was assigned to the respective segment. The mean ROI size was 685 ± 203 mm 2 . Large blood vessels, bile ducts and regions with partial volume, including air or perihepatic fat at the liver border, were excluded (Fig. 6). Another radiologist (A.H.), who was blinded to the patient's clinical history and had 7 years of experience in hepatic imaging, measured the shear modulus (in kPa) on MR elastography images and the PDFF (in %) based on in-and out-of-phase DIXON images in the right liver on three slices, using polygonal ROIs to exclude vessels and partial volume at the liver borders. The median value of the three ROIs was then calculated. statistical analysis. Analysis was performed with the statistical software package R (version 3.4.1, R Foundation for Statistical Computing) 38 and GraphPad Prism (version 7.1, GraphPad Software Inc.). Clinical characteristics were compared between groups using the Wilcoxon test for continuous variables or Fisher's exact test for categorical variables. The p-value for significance was <0.05. T 1 and T 2 * mapping parameters with good quality were then compared between segments using a paired Wilcoxon test. To address intersegmental variability and to identify the best area for reference, the median values of the liver segments adjacent to the lung (VII, VIII, IVa) and those not adjacent to the lung (VI, V, IVb, II, III) were calculated for the reference population and compared with the negative and positive controls using the Wilcoxon test.
To assess possible confounders in the reference population and among healthy volunteers (n = 79 + 12), a multivariate regression model was used with T 1 (segments not adjacent to the lung) as the outcome and sex (dummy-coded), age, PDFF and T 2 * (segments not adjacent to the lung) as variables. Age and sex were chosen as basic demographic characteristics that might bias the measured T 1 in the liver while PDFF and T 2 * were added as known representatives of liver fat and iron content that might influence T 1 . Pearson correlation was used to Figure 5. Study participant workflow. A total of 161 participants were recruited for the study. Six patients were excluded because of an aborted MRI scan due to claustrophobia (n = 1) or due to technically inadequate MRE scans (n = 5). Resulting included study population consisted of 12 healthy volunteers with normal liver stiffness and without steatosis (negative controls) and 143 patients without focal liver disease on CT. The MRI scans of 143 patients were included in statistical analysis. There were 102 patients who did not show steatosis, and 79 of these 102 patients also showed normal liver stiffness (reference patients). The remaining 23 showed increased liver stiffness (≥2.8 kPa). Forty-one patients had liver steatosis with PDFF > 10% (26 with normal liver stiffness, 15 with increased liver stiffness). Patients with increased liver stiffness and/or steatosis were defined as positive controls.