Feasibility and Reproducibility of BOLD and TOLD Measurements in the Liver With Oxygen and Carbogen Gas Challenge in Healthy Volunteers and Patients With Hepatocellular Carcinoma

Abstract

Purpose

To quantify baseline relaxation rates ${\mathrm{R}}_2^{\ast }$ and R₁ in the abdomen, their changes after respiratory challenges, and their reproducibility in healthy volunteers and patients with hepatocellular carcinoma (HCC) at 1.5T and 3.0T.

Materials and Methods

${\mathrm{R}}_2^{\ast }$ measurements were acquired in the liver in 8 volunteers and 27 patients with 34 HCCs using multiecho $T_2^{\ast }$ at baseline and after respiratory challenges with 100% oxygen (O₂) and carbogen (CB=95%O₂/5%CO₂). R₁ was measured at 1.5T in one volunteer and 21 patients with 23 HCCs. Test retest coefficient of variation (CV) was assessed in 10 subjects. Intra and interobserver variability of ${\mathrm{R}}_2^{\ast }$ and R₁ measurements was assessed in 12 and 10 patients, respec tively. Parameters for HCC, liver, and muscle were compared between baseline and after gas challenges.

Results

We observed that ${\mathrm{R}}_2^{\ast }$ and R₁ imaging of HCCs with O₂ and CB is feasible and reproducible (test retest CV ${\mathrm{R}}_2^{\ast }$<15%/R₁<5%; intra- and interobserver intraclass correlation coefficient ${\mathrm{R}}_2^{\ast }$>0.88/R₁>0.7 and CV ${\mathrm{R}}_2^{\ast }$<7%/R₁<3% at 1.5T). ${\mathrm{R}}_2^{\ast }$ measurements were observed to be less reproducible at 3.0T (CV<35%). There was a statistically significant decrease in ${\mathrm{R}}_2^{\ast }$ values in HCC before and after O₂ (P=0.02) and increase in R₁ after O₂ (P=0.004). CB had no significant effct (P ${\mathrm{R}}_2^{\ast }$=0.47/R₁=0.278).

Conclusion

${\mathrm{R}}_2^{\ast }$ measurements in HCC and liver parenchyma are more reproducible at 1.5T than at 3.0T, and with O₂ than with CB challenge. We observed a decrease in ${\mathrm{R}}_2^{\ast }$ and an increase in R₁ of HCCs from baseline in response to O₂ challenge, as expected with increased tissue and blood oxygenation.

Intratumoral hypoxia has been shown to correlate with tumor invasiveness, progression, and radioresistance in animal models and human carcinomas.¹ Tumor cells have been shown to adapt to hypoxic stress through genome and proteome changes induced via the hypoxia inducible factor 1-α (HIF 1-α) pathway, which promotes tumor progression and malignancy.^2,3 Moreover, low oxygen content in tumor tissue decreases the effectiveness of radiotherapy and chemotherapy.² Tumor oxygenation has been shown to be influenced by several factors including microvessel density, blood flow, blood volume, blood oxygen saturation, tissue pO₂, and oxygen consumption rate.⁴ Information on oxygen bioavailability in tumors can have important clinical implications, such as stratification of patients for therapy and monitoring response to systemic and locoregional therapy.^5,6

The development of new imaging techniques, such as oxygen-sensitive magnetic resonance imaging (MRI)^6–9 makes it possible to determine the perfusion and oxygenation state of tumors noninvasively. MRI techniques such as blood oxygen level-dependent (BOLD) and tissue oxygen level-dependent (TOLD)¹⁰ were first developed to study areas of activation in the brain,¹¹ but have since been used for in vivo evaluation of tumor blood supply and oxygenation.^{5–10,12–17}

In BOLD MRI, deoxygenated hemoglobin (deoxy-Hb) acts as an endogenous contrast agent by increasing magnetic susceptibility and shortening the transverse relaxation time ($T_2^{\ast }$) of the bulk magnetization of water protons in nearby tissues. In hypoxic tissues, the proportion of deoxy-Hb is greater, so the transverse relaxation rate ${\mathrm{R}}_2^{\ast }$ (1/$T_2^{\ast }$) is higher, and is expected to decrease during a hyperoxic respiratory challenge, as deoxy-Hb becomes more saturated with oxygen. Studies in rodents and humans have shown a BOLD signal increase (${\mathrm{R}}_2^{\ast }$ decrease) in various tumors, including hepatocellular carcinoma (HCC), in response to breathing 100% O₂^5,6,9 or carbogen (CB; 95% O₂ / 5% CO₂ or 98% O₂ / 2% CO₂).^7,8,10,16 Since Hb saturation with O₂ is dependent on the arterial O₂ pressure and oxygen partial pressure in tissue (pO₂), ${\mathrm{R}}_2^{\ast }$ can be thought of as an index of oxygenation. However, since other factors, such as blood volume, flow, and vessel geometry affect the concentration of Hb in tissue, ${\mathrm{R}}_2^{\ast }$ cannot be used to infer pO₂.

The TOLD effect has been observed in tumors and healthy tissues as an increase in the longitudinal relaxation rate of tissue magnetization (R₁=1/T₁) in response to O₂.^13,14 The TOLD effect relies on the paramagnetic properties of free molecular O₂, which acts as a weak T₁-shortening contrast agent.¹⁸ Since the concentration of free blood O₂ is substantial only in well-oxygenated tissues (when hemoglobin is over 90% saturated with O₂), changes in R₁ are sensitive to a different range of tissue oxygenation than changes in ${\mathrm{R}}_2^{\ast }$.⁷ A previous study of tumor xenografts in mice showed that BOLD and TOLD can better characterize tumor oxygenation in combination than each alone.⁷ BOLD or TOLD data in human HCC, at baseline or with physiological challenges, is very limited.

The purpose of our preliminary study was to quantify and assess the test–retest reproducibility, as well as the intra- and interobserver reproducibility at 1.5T and 3.0T, of baseline transverse (${\mathrm{R}}_2^{\ast }$) and longitudinal (R₁) relaxation rates, and their changes in response to 100% O₂ and CB respiratory challenges in HCC tumors, liver parenchyma, and muscle.

Materials and Methods

Patients

Between June 2013 to September 2014, patients with HCC and healthy volunteers were enrolled in our prospective, single-center, Health Insurance Portability and Accountability Act-compliant study approved by the Institutional Review Board (IRB) at our institution. Subjects with chronic obstructive pulmonary disease were not enrolled in the study, due to risk of respiratory decompensation during the hyperoxic or hyperoxic/hypercapnic respiratory challenges. Written informed consent was obtained from all subjects.

Eight healthy volunteers (M/F 1/7, mean age 36 years, range 23–55 years) were enrolled in the study for protocol optimization. Twenty-nine initial patients underwent multiparametric MRI of the liver with hyperoxic respiratory challenge. Two patients were excluded due to poor overall image quality (n=1) or lack of HCC on pathologic examination (n=1). Two patients had been treated with prior transarterial chemoembolization.

Finally, 27 patients (M/F 21/6, mean age 59 years, range 30–76 years) with resectable (n=15) and unresectable (n=12) HCC were included. HCC was diagnosed based on contrast-enhanced clinical MRI (n=26) or computed tomography (CT) (n=1) according to the American Association for Study of Liver Diseases (AASLD) 2011 criteria.¹⁹

Twenty-six patients had chronic liver diseases of various etiologies (chronic hepatitis C [n=18], chronic hepatitis B [n=6], NASH [n=2]) and one patient had no history of chronic liver disease (in this case, HCC diagnosis was confirmed on pathology).

Ten subjects underwent test–retest MRI examinations (seven patients and one volunteer at 1.5T, two patients at 3.0T) on two visits, up to 2 weeks apart (mean interval 6 days).

MRI

Subjects were imaged on a 1.5T (Aera, Siemens Healthcare, Erlangen, Germany) or a 3.0T (Skyra, Siemens Healthcare) system, each equipped with a 30 channel spine and flexible body array coil for radiofrequency (RF) receiving. All subjects were instructed to fast for 6 hours before the MRI exam. Six volunteers and one patient were imaged on both systems. Liver anatomy and the presence of lesions were evaluated in all subjects using axial and coronal T₂-weighted imaging (T₂WI) half-Fourier single-shot turbo spine-echo (HASTE), axial fat-suppressed T₂WI, axial dual-echo chemical shift imaging, and axial diffusion-weighted (DWI). 3D T₁-weighted imaging (T₁WI) pre- and postinjection of a gadolinium-based contrast agent, as well as DCE-MRI, was performed in all patients.

BOLD Imaging

A fat-suppressed, 2D multiecho gradient echo (MGRE) pulse sequence was acquired axially (Table 1) in all subjects, in 15-second breath-holds at end-expiration. At 1.5T, the protocol with five echoes in phase was initially chosen because in-phase conditions yield greater signal-to-noise ratio (SNR). However, since the MGRE sequence is fat-suppressed, fat-water chemical shift has a less pronounced effect on the signal, so we expanded the protocol to 12 echoes in a wider range of TEs (Table 1; 1.44–40 msec). The last TE (40 msec) was chosen to correspond to the $T_2^{\ast }$ of some HCCs. The first TE (1.44 msec) was chosen for greater sensitivity to decreased liver $T_2^{\ast }$ values in the presence of iron. ${\mathrm{R}}_2^{\ast }$ was measured in 11 patients and 7 volunteers using the five echoes protocol, and in 12 patients using the 12 echoes protocol. The 1.5T protocols were compared in seven subjects for room air respiratory conditions. For patients, slices were centered over the lesions identified on routine sequences. The MGRE sequence was performed before contrast administration at baseline and at the end of O₂ and CB breathing periods.

TABLE 1.

BOLD Imaging Protocol With a Multiecho Gradient-Echo (MGRE) Sequence and TOLD Imaging Protocol With an Inversion-Recovery Look-Locker (IR-LL) Sequence at 1.5T and 3.0T

	2D MGRE 5 echoes IP	2D MGRE 12 echoes	2D MGRE 7 echoes IP	2D IR -LL
Acquisition plane	Axial	Axial	Axial	Axial
Field	1.5T	1.5T	3.0T	1.5T
TR (msec)	242		165	2.26
TE (msec)	4.76, 9.53, 14.29, 19.06, 23.82	1.66,2.92,4.25,5.58,6.91, 10, 15, 20, 25, 30, 35, 40	2.46, 4.92, 7.38,9.84,12.3,14.76,17.22,	1.04
FA	35°	20°	35°	8°
TI (ms)	—	—	—	42-1576.5 32 TI’s every 48 msec
FOV read (mm)	340	340	340	420
FOV phase (%)	75	75	75	68.75
Slice thickness (mm)	7	7	7	8
Acquisition matrix	192 × 115	192 × 115		128 × 88
Slices	4-5	4-5	4-5	1-2
Parallel imaging factor	2	2	2
Acquisition time (sec)	15	15	15	20

TOLD Imaging

The TOLD effect was assessed by precontrast quantitative T₁ measurements at baseline and at the end of the gas challenges. A Look-Locker inversion recovery (IR-LL) pulse sequence with SNAPSHOT-FLASH readout²⁰ (Table 1) was used to provide quantitative T₁ mapping in 1–2 slices through each lesion in one 18-second breath-hold.

Respiratory Challenge

Hyperoxic (100% O₂) and hyperoxic-hypercapnic (CB) respiratory challenges were integrated into the multiparametric imaging protocol (Fig. 1). 100% O₂ available in the MRI exam room was administered at 15 L/min for 10–15 minutes, through a Hudson mask (Westmed, Responsive Respiratory, Tucson, AZ) with one-way non-rebreathing valves, covering the subject’s nose and mouth. CB (95% O₂ / 5% CO₂, TW Smith, Brooklyn, NY) was administered in the same way after a 2-minute recovery period under room air, during which the subject was breathing without the mask. BOLD and TOLD data were acquired during breath-holds at the end of each respiratory challenge. In all, 18 of 29 enrolled patients and all volunteers received both gas challenges, while 11 patients underwent only the O₂ challenge. CB challenge was not administered in the first three enrolled patients because we wanted to test the tolerance of volunteers to CB before administering it to patients. For subsequent patients, the CB challenge was not administered because of reduced respiratory capacity in one patient with pneumonectomy, patient’s refusal to undergo the CB challenge (n=2), and unavailability of CB (n=4). With both respiratory challenges we defined the failure rate as the percent difference between the number of HCCs imaged under the respective challenge with each technique, and the number of HCCs with interpretable images for the challenge.

FIGURE 1.

MRI acquisition protocol demonstrating the succession of respiratory challenges for BOLD and TOLD imaging.

Image Analysis

HCCs were identified by a single experienced radiologist (observer 1, C.B., with 5 years of experience) according to the AASLD 2011 criteria¹⁹ on standard imaging sequences and/or prior clinical imaging studies (Figs. (2 and 3)). An image analyst (observer 2, O.B.), a physicist with 6 years of experience in MRI, placed regions of interest (regions of interest [ROIs]: 15.2 ± 22.9 cm², range 0.9–83.2 cm²) on the MGRE and IR-LL images in consensus with observer 1. Images on which more than 50% of the lesion was not clearly visible due to motion artifacts, or obscured by tissue–air susceptibility artifacts, were considered noninterpretable and excluded from the analysis. Care was taken to contour the lesion cross-section of largest diameter for all respiratory conditions.

FIGURE 2.

A 67-year-old male patient with left hepatic lobe HCC measuring 5.4 cm (arrow on postcontrast T₁WI), which demonstrates a decrease in ${\mathrm{R}}_2^{\ast }$ after O₂ ($\Delta {\mathrm{R}}_2^{\ast }$ O₂=18.7%) and carbogen ($\Delta {\mathrm{R}}_2^{\ast }$ CB=20%) gas challenges. HCC shows no significant change in R₁ (ΔR₁ O₂=−1.7%), below the level of physiological variation denoted by test–retest CV=3.6%. The decrease in ${\mathrm{R}}_2^{\ast }$ above the level of physiological variation, accompanied by no significant change in R₁, is consistent with baseline hypoxia.

FIGURE 3.

A 68-year-old male patient with right hepatic lobe HCC treated with chemoembolization. There is central tumor necrosis on postcontrast T₁WI (arrow). The viable tumor at the periphery of the lesion demonstrates response to hyperoxia (arrows), with a decrease in ${\mathrm{R}}_2^{\ast }$ under O₂ ($\Delta {\mathrm{R}}_2^{\ast }$ O₂=17%) and carbogen ($\Delta {\mathrm{R}}_2^{\ast }$ CB=23.7%). The tumor also shows overall increase in R₁ under O₂ (ΔR₁ O₂=−8.5%). This lesion may be normoxic or mildly hypoxic, since the R₁ increase with O₂ challenge is also substantial.

Measurements were repeated in a sample of 12 patients for MGRE and 10 patients for IR-LL, by a third observer (M.W., a radiologist with 3 years of experience) to assess interobserver reproducibility. Observer 2 repeated the ${\mathrm{R}}_2^{\ast }$ and R₁ measurements in the same sample of 12 and 10 patients, respectively, for intraobserver variability assessment.

${\mathrm{R}}_2^{\ast }$₂ ANALYSIS

ROIs were delineated using Osirix (Pixmeo SARL, Switzerland) on the MGRE image with the longest TE, in the HCCs (encompassing the whole lesion, excluding necrotic areas, partial volume, and susceptibility artifacts), the right hepatic lobe (2 cm², avoiding artifacts and major blood vessels), and paravertebral muscles in the same slice as the lesion (${\mathrm{R}}_2^{\ast }$ values averaged for two ROIs of 2 cm², one on each side), for all respiratory conditions. As tissue–air susceptibility artifacts are most noticeable at the longest TE, ROIs were delineated on the image with longest TE to exclude all areas affected by artifacts from the ROI contour. Mean signal in the ROI was fitted to a monoexponential decay curve to measure $T_2^{\ast }$, using custom software written in MATLAB R2013b (MathWorks, Natick, MA). For lesions with maximum diameter greater than 6 cm, ${\mathrm{R}}_2^{\ast }$ was obtained as the average of ROI-based ${\mathrm{R}}_2^{\ast }$ values in two adjacent slices. ${\mathrm{R}}_2^{\ast }$ maps were calculated by fitting the signal pixel-by-pixel to the same monoexponential model.

R₁ ANALYSIS

T₁ maps were calculated in MATLAB from pixel-by-pixel fit of signal to the Look-Locker signal recovery equation.²⁰ ROIs were transferred in Osirix from the MGRE images to the IR-LL magnitude images and T₁ maps, with exclusion of partial volume artifact areas. R₁ was calculated as inverse of the mean T₁ value in each ROI. Changes in the transverse and longitudinal relaxation rate between baseline and hyperoxic respiratory challenges were calculated as:

$$\Delta {\mathrm{R}}_{1,2\ast (\%)}=100\ast [({\mathrm{R}}_{1,2\ast \text{baseline}}-{\mathrm{R}}_{1,2\ast \text{gas}})∕{\mathrm{R}}_{1,2\ast \text{baseline}}]$$

Statistical Analysis

Data are presented as mean ± standard deviation. Coefficient of variation (%) was used to assess test–retest reproducibility of ${\mathrm{R}}_2^{\ast }$ and R₁ in all studied tissues, on both platforms in a subset of 10 subjects. ${\mathrm{R}}_2^{\ast }$ values at 1.5T and 3.0T were compared by Mann– Whitney U-test. ${\mathrm{R}}_2^{\ast }$ and R₁ values for each tissue at room air and at hyperoxic respiratory challenges were compared using a paired Wilcoxon sign rank test. For assessment of lesion response to gas challenge, we used the CV at baseline to denote changes due to normal physiological variation. Lesions that showed absolute $\Delta {\mathrm{R}}_2^{\ast }$ or ΔR₁ with O₂ or CB greater than the mean test–retest CV were considered responsive to the increase in oxygen content. Intraobserver and interobserver reproducibility was assessed by measuring CV and intraclass correlation coefficients (ICC); ICC 0–0.2 indicates poor agreement: 0.3–0.4 indicates fair agreement; 0.5–0.6 indicates moderate agreement; 0.7–0.8 indicates strong agreement; and >0.8 indicates excellent agreement.²¹ Relationships between ${\mathrm{R}}_2^{\ast }$, R₁, and their changes with respiratory challenges to tumor size were assessed by Spearman correlation. ${\mathrm{R}}_2^{\ast }$, R₁, $\Delta {\mathrm{R}}_2^{\ast }$, and ΔR₁ for the different respiratory conditions were compared between lesions of different histopathologic grades by Kruskal–Wallis test, followed by Mann–Whitney U-test. $\Delta {\mathrm{R}}_2^{\ast }$ and ΔR₁ with O₂ and CB in HCC was compared to that of the liver and muscle by Mann–Whitney U-test.

Results

All subjects tolerated the multiparametric MRI exam, although all volunteers and patients who received CB reported greater difficulty in maintaining a breath-hold at the end of the 10-minute CB challenge. One patient refused to complete the CB challenge, and another refused a repeat of the same challenge in the retest exam. There were no adverse events reported with the use of gas challenges.

Forty-one lesions were identified in the 27 patients. Of the two treated patients, one had a 100% necrotic lesion that was excluded (while liver and muscle were analyzed) and the other one had a partially necrotic lesion (20% necrosis) that was analyzed (Fig. 3). One patient had two lesions, one HCC and a mixed HCC-cholangiocarcinoma on pathology; the latter was excluded from the study. In addition, 5 of 39 HCCs evaluated with R₂ under O₂ challenge were excluded due to motion and/or susceptibility artifacts (failure rate of 12.8%). Four of the 25 HCCs for which ${\mathrm{R}}_2^{\ast }$ imaging was attempted with CB were uninterpretable due to motion artifacts and were excluded (failure rate of 16%). Two of the 25 HCCs imaged with R₁ under O₂ did not have interpretable images due to motion artifacts and mismatch of tumor cross-sections between baseline and oxygen challenge (failure rate of 8%). With R₁ under CB, the failure rate was similar (15%), as 3 of 20 HCC exams were not evaluated due to motion artifacts.

Finally, 34 HCCs (mean size 4.3 ± 3.5 cm, range 1.0–13.0 cm) in 27 patients were evaluated for BOLD and 23 HCCs (mean size 4.2 ± 3.1 cm, range 1.4–13.0 cm) in 21 patients for both BOLD and TOLD effects (Table 2).

TABLE 2.

${\mathbf{R}}_2^{\ast }$ and R₁ Values at Baseline and During Respiratory Challenges, With Percentage Changes From Baseline

		HCC			Liver			Muscle
		n a	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$	n	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$	n	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$
All	Air	34	32.2±17.8		35	42.1±17.5		35	36.2±4.2
	O2	34	b29.5±10.6	4.7±10.3	35	40.9±19.4	3.3±14.1	35	36.6±4.2	21.4±8.7
	CB	19	31.7±16.28	3.1±18.8	24	40.8±16.1	0.03±13.3	21	b38.8±6.1	212.1±18.5
		na	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$	n	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$	n	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$
1.5T	Air	28	28.4±8.6		30	38.8±15.1		30	35.6±3.7
	O2	28	28.3±10.9	1.33±9.6	30	37.5±16.1	2.6±14.6	30	35.8±3.8	−0.8±5.7
	CB	18	29.6±13.5	−2.5±13.7	19	38.2±15.0	−2.03±13.3	20	38.2±6.0a	−11.7±19.3
		na	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$	n	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$	n	${\mathbf{R}}_2^{\ast }$	$\Delta {\mathbf{R}}_2^{\ast }$
3.0T	Air	8	45.7±14.6		12	61.2±14.9		12	40.9±3.8
	O2	8	42.8±16.3	7.9±13.9	12	61.3±19.5	1.28±10.6	12	40.7±3.5	−0.3±13.5
	CB	3	62.5±11.4	−15.9±22.3	9	60.2±18.9	−2.24±20.3	8	40.5±17.6	−12.3±26.9
		na	R1	ΔR1	n	R1	ΔR1	n	R1	ΔR1
1.5T	Air	23	1.34±0.2		21	1.76±0.2		22	1.38±0.1
	O2	23	1.39±0.2a	−3.82±5.4	21	1.7±60.2	−0.13±4.4	22	1.39±0.1a	1.39±0.1
	CB	16	1.37±0.2	−1.1±3.9	13	1.77±0.3	−0.74±4.7	14	1.41±0.1	1.41±0.1

n^a: number of lesions; n: number of patients;

^asignificant, paired Wilcoxon test (P < 0.05), for data grouped by field strength;

^bsignificant, paired Wilcoxon test (P < 0.05), for combined data at both field strengths.

$\Delta {\mathbf{R}}_2^{\ast }$ O₂ all HCCs vs. $\Delta {\mathbf{R}}_2^{\ast }$ O₂ liver, P= 0.884 (Mann-Whitney test).

$\Delta {\mathbf{R}}_2^{\ast }$ CB all HCCs vs. $\Delta {\mathbf{R}}_2^{\ast }$ CB liver, P= 0.452 (Mann-Whitney test).

$\Delta {\mathbf{R}}_2^{\ast }$ O₂ all HCCs vs. $\Delta {\mathbf{R}}_2^{\ast }$ O₂ muscle, P= 0.013 (Mann-Whitney test).

$\Delta {\mathbf{R}}_2^{\ast }$ CB all HCCs vs. $\Delta {\mathbf{R}}_2^{\ast }$ CB muscle, P = 0.538 (Mann-Whitney test).

ΔR₁ O₂ all HCCs vs. ΔR₁ O₂ liver, P= 0.0039 (Mann-Whitney test).

ΔR₁ CB all HCCs vs. ΔR₁ CB liver, P= 0.369 (Mann-Whitney test).

ΔR₁ O₂ all HCCs vs. ΔR₁ O₂ muscle, P= 0.024 (Mann-Whitney test).

ΔR₁ CB all HCCs vs. ΔR₁ CB muscle, P= 0.921 (Mann-Whitney test).

Liver and paravertebral muscles were analyzed for ${\mathrm{R}}_2^{\ast }$ and baseline and with O₂ in all of the 27 patients and in the eight volunteers (Table 2). The numbers of interpretable liver and muscle images with ${\mathrm{R}}_2^{\ast }$ under CB were smaller, due to motion artifacts (Table 2). R₁ was evaluated in the liver at baseline and under O₂ in 20 patients and one volunteer, and in the muscles, in 21 patients and one volunteer (Table 2). The number of subjects with interpretable R₁ images for liver and muscle images also decreased under CB due to motion artifacts (Table 2).

Histopathology

Histopathological proof of HCC was available for 18/34 lesions in 17 patients (partial hepatectomy, n=15, or biopsy, n=2), with the following tumor grades: well differentiated (two lesions), moderately differentiated (10 lesions), poorly differentiated (five lesions), undifferentiated (one lesion). All resected lesions had microvascular invasion.

Test–Retest Reproducibility

At 1.5T, ${\mathrm{R}}_2^{\ast }$ measurements for all studied tissues and all respiratory challenges had CV below 15% (Table 3). The test–retest CV for HCC was higher under CB (15%) than at baseline and under O₂ (9%), likely due to respiratory motion artifacts. Test–retest reproducibility was poorer at 3.0T, with mean CV for HCCs under O₂ at 21.5% and under CB at 32%. The two protocols used for quantification of ${\mathrm{R}}_2^{\ast }$ were equivalent in the liver, with bias of −1.3% and all values within the limits of agreement of [−7%, 4.4%]. In the eight HCCs identified in the patients used for protocol comparison, bias between the five echoes and the 12 echoes measurement was − 2.9%, and limits of agreement were [−19.2%, 13.5%]. R₁ (measured only at 1.5T) was highly reproducible, with CV <5% in all tissues, under all respiratory conditions (Table 3).

TABLE 3.

Mean Coefficients of Variation (%) for Test-Retest Variability Measurements of ${\mathrm{R}}_2^{\ast }$ and R₁ Values at Baseline and During Respiratory Challenges

	HCC			Liver		Muscle
1.5T	${\mathrm{R}}_2^{\ast }$
		n a	Mean CV (%)	n	Mean CV (%)	n	Mean CV (%)
	Air	7	9.1	8	11.64	8	7.6
	O2	7	9.43	8	12.09	8	12.9
	CB	2	15	4	8.7	4	5.9
	R1
		n a	Mean CV	n	Mean CV	n	Mean CV
	Air	8	3.67	7	4.12	7	1.61
	O2	8	4.23	7	4.06	7	1.1
	CB	2	3.58	2	4.37	3	3.23
3.0T	${\mathrm{R}}_2^{\ast }$
		n a	Mean CV	n	Mean CV	n	Mean CV
	Air	4	8.8	2	9.1	2	8.1
	O2	4	21.5	2	8.2	2	3.58
	CB	1	32.4	1	12	—	—

n^a: number of lesions assessed on test-retest; n: number of patients assessed on test-retest.

Intra- and Interobserver Reproducibility

${\mathrm{R}}_2^{\ast }$ had strong to excellent intra- and interobserver agreement in all tissues for all respiratory conditions (Table 4). R₁ measurements showed excellent intra- and interobserver agreement in all tissues for all respiratory challenges (Table 4).

TABLE 4.

Intra- and Interobserver Agreement for ${\mathrm{R}}_2^{\ast }$ and R₁ Measurements

		HCC (na = 16)			Liver (n=12)			Muscle (n=12)
${\mathrm{R}}_2^{\ast }$		Air	O2	CB	Air	O2	CB	Air	O2	CB
Intraobserver	CV (%)	2.55	2.25	3.92	3.18	2.72	11.15	1.91	1.08	3.05
	ICC	0.97	0.98	0.98	0.92	0.97	0.7	0.69	0.98	0.79
	P	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	0.0049	0.0048	<0.0001	<0.0001
Interobserver	CV (%)	4.86	3.66	6.19	3.62	6.06	10.31	0.86	1.93	2.66
	ICC	0.88	0.96	0.94	0.96	0.88	0.76	0.99	0.96	0.83
	P	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	0.0016	<0.0001	<0.0001	<0.0001
		HCC (na = 12)			Liver (n=10)			Muscle (n=10)
R1		Air	O2	CB	Air	O2	CB	Air	O2	CB
Intraobserver	CV (%)	1.75	1.89	2.35	1.33	1.61	1.99	1.87	1.44	2.14
	ICC	0.97	0.96	0.71	0.98	0.97	0.96	0.84	0.87	0.88
	P	<0.0001	<0.0001	0.009	<0.0001	<0.0001	<0.0001	0.0008	0.0002	0.0011
Interobserver	CV (%)	1.77	2.37	2.35	2.97	2.88	3.27	1.87	2.56	3.18
	ICC	0.96	0.95	0.85	0.84	0.87	0.91	0.79	0.8	0.77
	P	<0.0001	<0.0001	<0.001	0.0003	0.0003	0.0004	0.0025	0.0015	0.0079

n^a: number of lesions assessed; n: number of patients assessed.

Response to Hyperoxic Challenges

BOLD

Baseline ${\mathrm{R}}_2^{\ast }$ values were statistically lower at 1.5 T vs. 3.0T for HCC, liver, and muscle (P < 0.0001). We observed a significant decrease for all HCCs between baseline ${\mathrm{R}}_2^{\ast }$ and after O₂ challenge (paired Wilcoxon P=0.019), but not between baseline ${\mathrm{R}}_2^{\ast }$ and after CB (P=0.47) (Table 2). In the liver, we were unable to detect a statistically significant change in ${\mathrm{R}}_2^{\ast }$ with O₂, although there was a trend of decreasing ${\mathrm{R}}_2^{\ast }$ with O₂ (P=0.095). No significant change in ${\mathrm{R}}_2^{\ast }$ was observed with CB (P=0.71) in the liver (Table 2). Muscle ${\mathrm{R}}_2^{\ast }$ showed a statistically significant increase with CB (P=0.006), and no difference for O₂ (P=0.15). $\Delta {\mathrm{R}}_2^{\ast }$ O₂ observed in HCC was significantly greater than that observed in muscle, but not significantly greater than that observed in liver. There were no significant differences between HCC, liver, and muscle for $\Delta {\mathrm{R}}_2^{\ast }$ CB (Table 2).

Of note, the changes in ${\mathrm{R}}_2^{\ast }$ in HCC with O₂ and CB were 28 lesions studied with O₂ (baseline test–retest CV=9.1%) showed a decrease in ${\mathrm{R}}_2^{\ast }$ from baseline (mean $\Delta {\mathrm{R}}_2^{\ast }$ O₂=16.8%), one showed an increase (D${\mathrm{R}}_2^{\ast }$ O₂=−27.7%), and the majority (23 out of 28, $\Delta {\mathrm{R}}_2^{\ast }$ O₂=4.3%) were considered nonresponders (Fig. 4). With CB (Fig. 4), 4 out of 18 lesions showed a decrease in ${\mathrm{R}}_2^{\ast }$ ($\Delta {\mathrm{R}}_2^{\ast }$ CB=17.6%), five showed an increase ($\Delta {\mathrm{R}}_2^{\ast }$ CB=−20.3%), and nine were considered nonresponders ($\Delta {\mathrm{R}}_2^{\ast }$ CB=5.1%). Of the eight lesions studied at 3.0T with O₂, four showed a decrease in ${\mathrm{R}}_2^{\ast }$ ($\Delta {\mathrm{R}}_2^{\ast }$ O2=18%), one showed an increase ($\Delta {\mathrm{R}}_2^{\ast }$ O2=−17.54%), and three were nonresponders (D${\mathrm{R}}_2^{\ast }$ O2=5.4%). With CB at 3.0T, two lesions showed a decrease in ${\mathrm{R}}_2^{\ast }$($\Delta {\mathrm{R}}_2^{\ast }$ CB=−25.3%), and one showed no response ($\Delta {\mathrm{R}}_2^{\ast }$ CB=2.8%).

FIGURE 4.

Histogram distribution of changes in ${\mathrm{R}}_2^{\ast }$ observed in HCC lesions at 3.0T (patients 1–5) and 1.5T (patients 4–27) with oxygen (O₂) and carbogen (CB) gas challenges. The horizontal lines mark the threshold for response ±9% (1.5T CV ${\mathrm{R}}_2^{\ast }$ baseline=±9.1%, 3.0T CV ${\mathrm{R}}_2^{\ast }$ baseline=±8.8%). $\Delta {\mathrm{R}}_2^{\ast }$ >0 denotes decrease in ${\mathrm{R}}_2^{\ast }$ from baseline with respiratory challenge ($\Delta {\mathrm{R}}_2^{\ast }$ (%)=100*[(${\mathrm{R}}_2^{\ast }$ baseline − ${\mathrm{R}}_2^{\ast }$ gas)/${\mathrm{R}}_2^{\ast }$ baseline). Lesions are indexed by patient number and ordered by superior-to-inferior anatomical location. The changes in ${\mathrm{R}}_2^{\ast }$ with O₂ are highly variable between individual lesions. With CB, 6/18 lesions at 1.5T and 2/3 lesions at 3T have paradoxical response (increase in ${\mathrm{R}}_2^{\ast }$, $\Delta {\mathrm{R}}_2^{\ast }$ <0).

There was no correlation between baseline ${\mathrm{R}}_2^{\ast }$ in HCCs and tumor size (Spearman’s r=−0.174, P=0.33), while there was a borderline significant correlation between R₂ O₂ and tumor size (Spearman’s r=−0.32, P=0.06), and a moderate significant correlation between $\Delta {\mathrm{R}}_2^{\ast }$ O₂ and tumor size (Spearman’s r=0.54, P=0.0001). For ${\mathrm{R}}_2^{\ast }$ under CB challenge, borderline significant Spearman correlations with lesion size were obtained (${\mathrm{R}}_2^{\ast }$ CB Spearman’s r=−0.44, P=0.057; $\Delta {\mathrm{R}}_2^{\ast }$ CB Spearman’s r=0.44, P=0.059).

TOLD

R₁ in the HCCs showed a significant increase with O₂ (paired Wilcoxon P< 0.001), which is expected with the increase of free O₂ concentration in the tissue (Table 2). The same increase in R₁ was not reproduced with CB (P=0.278). No statistically significant changes in R₁ were observed for the liver (O₂/CB P=0.79/0.89). A statistically significant increase in R₁ was observed for muscle with O₂ (P=0.0495), but not with CB (P=0.28). ΔR₁ O₂ for all HCCs (Table 2) was statistically greater than ΔR₁ O₂ in the liver (Mann–Whitney P=0.0039), and ΔR₁ O₂ in muscle (P=0.024).

An increase in R₁ with O₂ administration (Fig. 5) was observed in 10 out of 23 lesions (ΔR₁ O₂=−8.5%), while a decrease was observed in one lesion (undifferentiated HCC) (ΔR₁ O₂=10%), while 12 lesions showed no difference (ΔR₁ O₂=1.2%).

FIGURE 5.

Histogram distribution of changes in R₁ observed in HCC lesions at 1.5T, with oxygen (O₂) and carbogen (CB) gas challenges. The horizontal lines mark the threshold for response, CV R₁ baseline=3.6%; ΔR₁<0 denotes increase in R₁ from baseline with respiratory challenge (ΔR₁ (%)=100 *[(R_1baseline − R_{1 gas})/R_{1 baseline}). Lesions are indexed by patient number and ordered by superior-to-inferior anatomical location. The majority of lesions show the expected response with O₂, increase in R₁ (ΔR₁<0), while response with CB is variable.

Administration of CB resulted in no R₁ response in 14 out of 16 lesions (ΔR₁ CB=2.1%), while an increase or decrease in R₁ were observed in one and one lesion, respectively.

Significant negative correlations were observed between R₁ at baseline (Spearman’s r=−0.52, P=0.01), under O₂ (Spearman’s r=−0.48, P=0.022), and under CB (Spearman’s r=−0.50, P=0.046) with lesion size. No statistically significant correlations were observed between ΔR₁ O₂ or ΔR₁ CB and HCC size (Spearman’s r=0.22, P=0.312; Spearman’s r=0.22, P=0.31, respectively).

Combined BOLD and TOLD Findings

Five out of 22 lesions which had both ${\mathrm{R}}_2^{\ast }$ and R₁ measurements (Fig. 6) showed a substantial decrease in ${\mathrm{R}}_2^{\ast }$ (above test–retest CV) and modest response in R₁ (below test–retest CV) with O₂. The same response pattern was reproduced with CB in one of the five tumors (lesion 11-1). Only three lesions showed a substantial increase in R₁ and modest decrease in ${\mathrm{R}}_2^{\ast }$ with O₂ (Fig. 6).

FIGURE 6.

Histogram distribution comparing ${\mathrm{R}}_2^{\ast }$ and R₁ response in HCCs to oxygen (O₂) challenge in 22 HCCs. Some tumors show substantial decrease in ${\mathrm{R}}_2^{\ast }$ with oxygen from baseline, and modest/no increase in R₁; these are possibly hypoxic (thin arrows). Other tumors show a modest decrease in ${\mathrm{R}}_2^{\ast }$ and substantial increase in R₁ (thick arrows); these are possibly normoxic/mildly hypoxic/hyperoxic tumors.

Correlation With Histopathology

None of the ${\mathrm{R}}_2^{\ast }$ or R₁ measurements were significantly different between different grades of tumor (Kruskal–Wallis test P-value range 0.117–0.857).

Discussion

BOLD and TOLD have emerged as noninvasive alternatives to characterize tumor oxygenation.^{5,9,11,22–24} Although their noninvasive nature and wide availability make BOLD and TOLD attractive for clinical use, proof of reproducibility is necessary before their use as imaging biomarkers. Our study expands the scope of previous work on BOLD and/or TOLD in liver tumors^8,12,14,24 by examining both effects in human HCC with O₂ and CB challenge. Our experience shows that ${\mathrm{R}}_2^{\ast }$ measurements with respiratory challenge are feasible in human HCC and more reproducible at 1.5T than at 3.0T, and that R₁ measurement is highly reproducible at 1.5T. Although we observed a significant decrease in ${\mathrm{R}}_2^{\ast }$ and increase in R₁ of HCC with O₂ challenge, ${\mathrm{R}}_2^{\ast }$ and R₁ responses to O₂ and CB challenges were highly heterogeneous, possibly due to variability in blood flow / blood volume and capacity for vasodilation among tumors.

${\mathrm{R}}_2^{\ast }$ measurement in the liver is particularly challenging because of artifacts due to respiratory motion, fluctuations of the magnetic field at tissue–air interfaces, chemical shift, and cardiac pulsation.⁷ To avoid the documented effects of increasing glucose levels on blood flow²⁵ and ${\mathrm{R}}_2^{\ast }$ in the liver,²⁶ all subjects were imaged in fasting conditions. We obtained higher variability in ${\mathrm{R}}_2^{\ast }$ measurements at 3.0T than at 1.5T, which is consistent with greater magnetic field inhomogeneity at 3.0T. To control for the effect of chemical shift on signal, we used fat saturation in both our five in-phase echoes and 12 echoes protocol. ${\mathrm{R}}_2^{\ast }$ values from both protocols were found to be equivalent and in accordance with previously published ${\mathrm{R}}_2^{\ast }$ values for liver and HCC ${\mathrm{R}}_2^{\ast }$ at 1.5T^7,12 and 3.0T.²⁶ For T₁ measurement in the liver, we sacrificed spatial coverage for more accurate T₁ quantification with a more time-intensive Look-Locker method, which acquires many data points to sample the longitudinal relaxation of signal. We found that T₁ measurement was highly reproducible on test–retest measurements, and repeated observations by two observers. T₁ values obtained in the liver and muscle were within the expected published range.^13,14 Variable flip angle (VFA) methods can provide full liver coverage in a short time; however, quantification of T₁ is less accurate due to B₁ field inhomogeneities, particularly at higher field strength.^5,27

We observed a significant decrease in ${\mathrm{R}}_2^{\ast }$ of HCCs from baseline with administration of O₂, as expected. However, since deoxy-Hb saturation depends on blood volume, flow, and vessel geometry, $\Delta {\mathrm{R}}_2^{\ast }$ is influenced by both lesion pO₂ and vascularity, with changes in blood volume introducing perturbations in BOLD signal.⁴ Correlation with measures of vascularity, such as DCE-MRI, is needed to differentiate hypoxic from hypovascular tumors.⁴ Contrary to published BOLD data in other tumors,^{5,7–9,16,17} we did not observe a statistically significant decrease in ${\mathrm{R}}_2^{\ast }$ with the administration of CB. Some lesions showed a paradoxical response, an increase in ${\mathrm{R}}_2^{\ast }$ with administration of O₂ or CB, which has also been observed by Jhaveri et al¹² in their study of human HCC with O₂ challenge and by Guo et al⁸ in a study of the Novikoff hepatoma model in rats, with CB. The increase in ${\mathrm{R}}_2^{\ast }$ with CB can be explained by the immature vasculature in some HCCs, which fail to dilate in response to CO₂.²³ Hypercapnia causes greater vasodilation in the mature vasculature of neighboring liver tissue, which in turn causes vascular steal away from the tumor, and thus poorer oxygen delivery.⁸ Hypercapnia can also decrease Hb affinity for O₂, thus increasing ${\mathrm{R}}_2^{\ast }$.⁴

We observed a significant increase in R₁ of HCC in response to O₂. Although an increase in R₁ has been observed in response to O₂ and CB in other body carcinomas,^5,7 there are scarce data on HCC. The only study of the TOLD effect in 10 patients with abdominal tumors¹⁴ showed a statistically significant increase in R₁ with O₂, based on multiple measurements in one patient with HCC. A decrease of R₁ with O₂ or CB administration, observed in some HCCs in our study, is possible whenever O₂ delivery to the tumor tissue is impaired, such that the concentration of free paramagnetic O₂ is smaller. This may be due to vascular steal, or to a local increase in deoxy-Hb concentration due to hypercapnic vasodilation.¹⁰

In 5 out of 22 lesions that had both ${\mathrm{R}}_2^{\ast }$ and R₁ measurements with O₂, and in one of five lesions that had ${\mathrm{R}}_2^{\ast }$ and R₁ measurements under CB, a substantial decrease in ${\mathrm{R}}_2^{\ast }$ was associated with a modest response in R₁, a response observed previously in hypoxic tumors.⁷ Only three lesions showed a pattern of response associated with baseline normoxia/mild hypoxia or hyperoxia,⁷ namely, a substantial increase in R₁ and modest decrease in ${\mathrm{R}}_2^{\ast }$ with O₂.

The liver and paravertebral muscle did not show a change in ${\mathrm{R}}_2^{\ast }$ in response to O₂ inhalation, in accordance with previous studies.¹³ However, with CB inhalation we were able to reproduce the results of O’Connor et al, showing a significant ${\mathrm{R}}_2^{\ast }$ increase in the paravertebral muscles, but we did not reproduce the significant increase in ${\mathrm{R}}_2^{\ast }$ with CB in the liver, or the significant increases in R₁ in the liver and muscle with both O₂ and CB.¹³ The discrepancy may be due to different methods for ${\mathrm{R}}_2^{\ast }$ and R₁ quantification: O’Connor et al¹³ used an MGRE sequence with two TEs for ${\mathrm{R}}_2^{\ast }$ measurement, and a dual flip angle 3D SPGR sequence for T₁ measurement, which may not be as sensitive as our methods.

We administered CB after a 2-minute recovery period, during which the patient was breathing room air. Although other investigators used longer recovery periods (8 minutes),¹³ we judged 2 minutes to be sufficient based on the work of Remmele et al in brain tumors,¹⁰ which showed ${\mathrm{R}}_2^{\ast }$ and R₁ to revert to baseline values 100 seconds after the end of a hyperoxic challenge. We found that the CB challenge did not produce a more significant response in the ${\mathrm{R}}_2^{\ast }$ or R₁ of HCCs than O₂. Four out of 19 HCCs imaged with both gases showed the same type of ${\mathrm{R}}_2^{\ast }$ response under O₂ and CB (ie, an increase or decrease greater than physiological variation), and in these cases the $\Delta {\mathrm{R}}_2^{\ast }$ was greater with CB. Furthermore, administration of CB, although tolerated by patients, introduced more respiratory motion artifacts. The artifacts account for the failure rate of imaging with CB in our study (comparable to the ${\mathrm{R}}_2^{\ast }$ failure rate observed with CB in nonabdominal human tumors¹⁵, and for higher test–retest variability. A possible solution to the decreased failure rate of MRI with CB challenge would be to use a 98% O₂ / 2% CO₂ mixture instead of a 95% O₂ / 5% CO₂ mixture.¹⁵ More sophisticated physiological monitoring of end-tidal CO₂ pressure during CB challenge would also allow finer calibration of the hyperoxic-hypercapnic challenge and monitoring of physiological response.²⁸

$\Delta {\mathrm{R}}_2^{\ast }$ under O₂ and CB challenge were found to have modest positive correlation (significant and borderline significant, respectively) with lesion size. The borderline significant correlations for ${\mathrm{R}}_2^{\ast }$ O₂ and ${\mathrm{R}}_2^{\ast }$ CB with lesion size were negative, however. Our finding contradicts a previous study in rat HCC, which found significant negative correlation between $\Delta {\mathrm{R}}_2^{\ast }$ CB and lesion size, with $\Delta {\mathrm{R}}_2^{\ast }$ close to null or even negative for larger lesions. The discrepancy may be explained by the larger number of tumors (n=34) with greater variation in etiology in our human study, versus the small number of lesions (n=14) of controlled etiology in the rodent study.⁸

We found significant negative correlations between R₁ at baseline, R₁ O₂ and R₁ CB, and lesion size. Lower R₁ at baseline and in response to O₂ and CB observed with increasing lesion size suggests poor oxygenation at baseline⁷ in larger HCCs, possibly due to impaired O₂ delivery.

Our study had several limitations. The results summarize our initial experience with BOLD, TOLD, O₂, and CB challenge in a small series of subjects recruited in the first year of a 5-year prospective study. To assess the BOLD and TOLD effects, we used an imaging protocol with breathhold exams, similar to protocols used by other investigators in the abdomen.¹² Use of a breath-hold at end expiration may modify oxygen demand and availability. Breath-holds after CB breathing also decreased subjects’ comfort and cooperation, and caused a large number of motion artifacts on the images. Furthermore, the voxel size and number of slices was not identical between the BOLD and TOLD sequences, which precluded exact registration of ROIs between the BOLD and TOLD series. A single reader identified the HCCs based on clinical and research imaging exams. Since not all lesions studied were resectable, we had limited pathology confirmation of HCCs (available in 17/27 patients).

A future study could correlate ${\mathrm{R}}_2^{\ast }$ and R₁ results with DCE-MRI metrics of perfusion and with histopathology markers of vascularity and hypoxia, such as microvessel density and HIF 1-α expression.^7,29

We conclude that ${\mathrm{R}}_2^{\ast }$ and R₁ measurements in the liver and HCCs with O₂ and CB challenge are feasible, with better test–retest reproducibility of ${\mathrm{R}}_2^{\ast }$ measurements in HCC at 1.5T, and with O₂ compared to CB challenge. Our preliminary experience with BOLD and TOLD MRI demonstrates the variable response of HCC to O₂ and CB challenges, which should be correlated with DCE-MRI findings to decouple tumor oxygenation from blood flow and volume effects, and with advanced histopathology findings.