Is ADC established as an imaging biomarker for stereotactic body radiation therapy assessment in hepatocellular carcinoma?

Abstract

In this article, as part of this special issue on biomarkers of early response, we review currently available reports regarding magnetic resonance imaging apparent diffusion coefficient (ADC) changes in hepatocellular carcinoma (HCC) in response to stereotactic body radiation therapy. We compare diffusion image acquisition, ADC analysis, methods for HCC response assessment, and statistical methods for prediction of local tumor progression by ADC metrics. We discuss pros and cons of these studies. Following detailed analyses of existing investigations, we cannot conclude that ADC is established as an imaging biomarker for SBRT assessment in HCC.

Introduction

Stereotactic body radiation therapy (SBRT) is a safe and effective treatment for patients with hepatocellular carcinoma (HCC) and good liver function (Child-Pugh (CP) A) and is able to achieve local tumor control rates of 99–95% for 1-year and 95%−89% for 2-years^1,2. In addition, a recent clinical trial showed that the survival of patients with HCC receiving SBRT plus systemic therapy is superior to that of those receiving systemic therapy alone³. However, liver function decline post SBRT, particularly for the patients with poor liver function, is still a concern. A study showed that 2-point Child-Pugh (CP) score worsening at 6 months post SBRT could be reduced to 7% using an adaptive treatment strategy based upon individual patient liver function tolerance, compared to 27% without adaptation^1,2. To effectively treat the patients with HCC by SBRT, radiation dose-based tumor control probability (TCP) models have been developed². Imaging biomarkers could provide individual patient sensitivity and response to doses, and thereby help to improve the prediction of tumor control. With an increase of the availability of MRI-Linacs, HCC response to SBRT can be assessed at high frequency, e.g., at each faction of SBRT using MRI⁴. MRI diffusion imaging is an attractive candidate as it can be acquired within a few minutes and does not require any contrast agent. This article will review the diffusion imaging techniques and reported investigations on diffusion imaging as a biomarker for assessment of SBRT response in HCC. Although we only include the studies focused on HCC, we feel that the concepts will also apply to the radiation response of other diseases but need clinical evaluations in each disease. We intend to focus on the investigations of SBRT, considering that the HCC characteristics post-SBRT could be different from post transarterial chemoembolization, transarterial radioembolization or thermal ablation^5,6.

Diffusion imaging techniques

Diffusion weighted (DW) imaging measures the random motion of water molecules within tissue using diffusion-sensitizing gradients. Factors that influence diffusion sensitivity include water diffusion as well as cellular structure, cell membrane permeability and cytoplasmic organelle. The strength and duration of the diffusion-sensitizing gradients can be quantified by a single metric called b-value. DW imaging in the liver faces another challenge, respiratory motion, which can cause motion artifact, blur, and lower signal-to-noise ratio (SNR) in images. To manage breathing motion effects on DW images, several strategies have been used during acquisition, such as breath-holding, respiratory triggering acquisition, and free-breathing with a large amount of signal averaging. The free-breathing acquisition, although having less requirements of patient cooperation, often results in blur and low-quality diffusion weighted images, compared to breath-holding and respiratory retriggering. Breath-holding acquisition requires the most patient cooperation and often needs multiple breath-holdings due to the duration of acquisition⁵. The respiratory triggering acquisition requires a patient breathing normally but prolongs the total acquisition time due to only a small portion of the breath-cycle used for data acquisition.

Apparent diffusion coefficient (ADC) is a commonly used metric and can be calculated from DW images measured at minimal two b-values by fitting a mono-exponential decay function. The two b-values are often chosen are “zero” and “non-zero”. The liver is a highly perfused organ, in which perfusion causes pseudo-diffusion signals and increases ADC values. The perfusion-caused pseudo-diffusion can be identified as exhibiting a deviation of diffusion weighted signals from a mono-exponential decay at zero and low b-values^7,8. To separate true diffusion signals from perfusion-caused pseudo-diffusion, an intravoxel incoherent motion (IVIM) model was proposed by Le Bihan⁷, which requires the acquisition of DW images at multiple b-values to fit multiple parameters. Robustness of the fitted parameters is challenged by the low SNR and the limited number of the b-values acquired in DW images. An alternative approach to eliminate the effect of perfusion on the quantified ADC values is to use the DW images at b-values following the mono-exponential decay, for example at b-values greater than 200–300 s/mm².⁸

ADC changes in HCC in response to SBRT

Although diffusion weighted MRI techniques were introduced to the radiology clinic more than three decades ago, we could find only four studies in the literature, including retrospective analyses, that reported associations of ADC response in HCC post or during SBRT with local tumor progression or RECIST defined response^9–12. Loosening our criteria, we include one prospective pilot study in which diffusion imaging was performed from pre- to during and to post SBRT in patients but only 36% had HCC¹³ and another prospective study that reported the association of the ADC pre fractionated RT in HCC with local tumor progression risk¹⁴. The numbers of patients, DW acquisitions and analyses, clinical endpoints, statistical methods of the six studies are listed in Table 1. In the following paragraphs, we will discuss what we can learn from these studies.

Table 1.

Summary of Studies

study	No. pts/Tumor type/median size	therapy	Scanner/Acquisition BH,FB,RT/b-value s/mm2/Mono-exp or IVIM	Scan Time Pre/post	Analysis/Endpoint/model	Findings
Eccles(2009) Prospective	11/HCC(4), mets(5), cholangio(2)	SBRT: median 43.9 Gy (28.8–54Gy)/6 Fx	1.5T/BH/0,600/Mono-exp	Pre/during/1M post	ROI on ADC/RECIST defined response/Pearson correlation	ADC changes during SBRT were significantly associated with responses.
Lo (2019)retrospective	34/HCC/3.9cm (0.88–22)	SBRT: 30–60 Gy/4–6 Fx	1.5T/BT/0, 500/Mono-exp	Pre/6M post	3ROIs on ADC/mRECIST defined response/Cox model	Pre- or post-SBRT ADC values and DADC% were not significantly different bwt responders and non-responders. ADC%>25% was significant for local tumor control by Cox
Yu(2014)retrospective	48/HCC/<3cm	FRT(63%)+SBRT(37%)SBRT: 15–20Gy/3Fx	3T/NI/100, 800/Mono-exp	Pre and 3–5M post	ROI for ADC/mRECIST defined LPFS/KM for LPFS	univariate analysis of LPFS: 1) mRECIST(p<0.001). 2) size (p<0.01); 3) ADC% >20% p<0.02.
Song(2019)retrospective	39/HCC	SBRT:35–50Gy/10 Fx	3T/NI/NI/NI	Pre and within3M post SBRT	ROI for ADC/mRECIST defined response at 6M and Time-to-Progression/exact test; Cox model	Qualitive assessment on DWI post-RT was significant different between responders and non-responders (p<0.01); But ADC% was not. ADC or ADC% did not predict TTP.
Oldrini (2017)retrospective	27/HCC/2cm(<6 cm)	SBRT: 45 Gy/3 Fx	1.5T/NI/600/NA	Pre, and 3- and 6- month post	Hyperintesnity on DWI/RECIST and mRECIST defined response/Cox model	Absent hyperintensity signals from DW images pre and post SBRT were predictors of better response but weaker than tumor size.
Kim (2014)prospective	20/HCC/11cm	FRT w 5-FU: 45Gy/25Fx	3T/FB/50,400,800/mono-exp	Pre RT	ROI on ADC/RECIST define response & PFS/KM	Higher ADC pre-RT had significantly longer PFS than those with lower ADC.

BH=breath-hold; FB=free-breathing; RT=respiratory triggering; NI=no information

The Eccles’ study is the only prospective study having longitudinal diffusion scans pre, during and post SBRT, and can provide us a temporal profile of ADC changes in response to SBRT in the liver cancers¹³. This study enrolled 11 patients, 4 with HCC, 5 with metastases to liver, and 2 with cholangiocarcinoma, treated by a median dose of 43.9 Gy (28.8–54 Gy) in 6 fractions over 2 weeks. The patients had 4 diffusion MRI scans at pre, during the 1^st week (between 1 and 3 fractions), during the 2^nd week (4–6 fractions), and 1-month post SBRT. Diffusion weighted images were acquired on a 1.5 T scanner at b-values of 0 and 600 s/mm² using breath-holding. Subsequently ADC maps were computed using a mono-exponential decay. The ADC analysis was on a single region of interest (ROI) with a size of 1 cm and placed in the high dose region. The mean ADC value was 1.56×10⁻³ mm²/s pre-SBRT, increased by 21.2% at the 1^st week during SBRT, by 22.4% at the 2^nd week during SBRT and by 28.8% 1-month post-SBRT. The ADC% at the 1^st week was correlated with RECIST defined responses. This study suggested that the ADC was increased in response to the 1^st week radiation doses, but the subsequent increases in response to additional doses during 2^nd week therapy and the time interval of 1-month post SBRT were small. However, the interpretation of the finding of the correlation of the ADC changes at the 1^st week with HCC tumor responses should be made with caution due to the inclusion of mixed patient cancer types and the small number of the patients in this study. Furthermore, the two patients with cholangiocarcinoma had the smallest ADC increase and worse local control which could be dominate the correlation findings. Having said that, the possibility that ADC could be an early response marker for cholangiocarcinoma should be explored.

The four studies that reported ADC responses in patients with HCC from pre to 3–6 months post SBRT were all retrospective and correlated ADC changes with either responses or time-to-progression by various statistical analyses^9–12; see a summary in Table 1. The post-SBRT responses were defined by modified RECIST (mRECIST) or RECIST. Diffusion imaging acquisitions seem to be have done with free-breathing. The ADCs were calculated by fitting the mono-exponential decay to the diffusion weighted images (except Oldrini et al, 2017 in which the analysis was performed on DW images but not on the ADC). All analyses of ADC or DW images were performed in one or three regions of interest (ROI) placed on the tumors rather than within a tumor volume, which is subject to large inter-observer selection variations. The ADC values pre and post SBRT and ΔADC% of these studies were summarized in Table 2. As we have discussed, the studies that used diffusion weighted images with b-value of zero for ADC calculation reported the larger ADC values than those that did not. Overall, the association of the ADC change with tumor local control had mixed results. In the following paragraphs, we will discuss major findings of the four studies.

Table 2.

ADC pre and post SBRT and percentage changes in ADC in literature

study	b-values (s/mm2)	ADC (x10−3 mm2/s)			ΔADC% post-SBRT
		Pre SBRT	During SBRT	1–6M post SBRT	whole	responder	Non-responder
Eccles	0,600	1.56	1.89	2.01	28.8
Lo	0,500	1.43		1.72	20.3	27.6%	13.3%
Yu	100,800	1.21		1.41	16.5
Song	NI	1.02		1.43*	40.3*	46.7	21.9
Kim	50,400,800	1.05
ours	50,400,700,1000	1.14		1.29	14.3
ours	400,700,1000	1.00		1.15	16.4

Lo et al reported that ADC values pre- and 6-month post-SBRT and ADC% were not significantly different between responders and non-responders defined by mRECIST⁹. Then, the investigators looked for an optimal threshold value of ADC changes using ROC analysis. A cut-off value of 25% increase of ADC was found to be a significant predictor for time to tumor local progression by the Cox proportional-hazards model. Meanwhile, they found that the tumor size change from pre to post SBRT was a much stronger predictor than the ADC change. It is not clear whether the ADC change as a continual variable is a predictor or not in the Cox model of time to tumor local progression. Yu et al investigated ADC percentage increases 3–5 months post SBRT for prediction of time to local tumor progression defined by mRECIST using the Kaplan-Meier (KM) method and found that a threshold, 20% of ADC change, was a significant predictor¹². Song et al reported that ADC values pre or post SBRT and percentage changes in ADC of 39 patients with HCC were not significantly different between responders and non-responders at 6-month post SBRT defined by mRECIST and did not predict time to local tumor progression¹¹. But they reported that qualitive assessments of signals on DW images post-SBRT were significantly different between responders and non-responders. It is not very clear how this qualitive assessment was done and whether the finding is reproducible. Oldrini et al analyzed hyperintensity signals on diffusion weighted images acquired at b-value of 600 s/mm² pre and post SBRT in 27 patients with HCC¹⁰. They found that absent hyperintensity signals on the DW images pre and post SBRT both were predictors for better responses, but the predictions were weaker than ones using tumor sizes pre and post SBRT. The hyperintensity signals on DW images in HCCs generally indicate low ADC values and restricted diffusion in HCC compared to surrounding liver. Song et al reported that a large fraction of patients (30–40%) had no restricted diffusion in HCC pre-SBRT, and having no restricted diffusion in HCC pre and post SBRT was associated with better responses¹¹. The pre-SBRT results of Song’s study are consistent with Kim’s study in which they found that higher ADC values pre-RT in HCC were significantly associated with superior progression free survival¹⁴. However, it is not clear whether temporal profiles of ADC changes in response to SBRT in HCCs with less restricted diffusion were different from ones with restricted diffusion, and whether stratifying HCCs based upon restricted diffusion pre-SBRT could improve the predictive power of ADC changes post-SBRT for local tumor progression risk.

There are a few of weakness common to the four studies^9–12,14. First, they are retrospective. Second, they used ROI-based analyses, with three using a single ROI and one using 3 ROIs. The ROI definition is subject to large inter-observer variations and hard to reproduce. Third, most studies used free breathing during diffusion image acquisition, which could cause low quality and low signal-to-noise ratio in diffusion images. Fourth, there was lack of appropriate mitigation of pseudo-diffusion caused perfusion. The two studies including zero b-value for the ADC calculation have greater average ADC values (1.43–1.56 ×10⁻³ mm²/s) than those excluding zero b-value (1.02–1.21 ×10⁻³ mm²/s) (see Table 2). For demonstration, we include ADC values pre and 1-month post SBRT as well as ADC changes of 47 HCCs from 39 patients treated by 3–5 fractions of SBRT from our prospective study in Table 2. We show that ADC values including the DW images acquired at a low b-value of 50 s/mm² in the calculation are greater than ones excluding the images with low b-values. Fifth, there are large variations among studies (40% to 16%) in ADC percentage changes post-SBRT. This large variation cannot be completely attributed to the time variations of post-SBRT imaging since Eccles’ study showed the further ADC increases from the end of SBRT to 1-month post SBRT were small. It is not clear how much these variations are due to ROI placements. Sixth, there were variations on how to define response, in which both RECIST and modified RECIST criteria have been used. The two criteria seem to result in different response rates. Although the impact of all these variations on the predictive power of post-SBRT ADC changes on local tumor progression risk is hard to quantify, these variations likely affect model reproducibility. Most importantly in the context of this issue of The Cancer Journal, it is not clear whether ADC increases in HCC post-SBRT are specific enough for local tumor progression risk assessment.

Diffusion-based Cellular Imaging

As discussed, multi-factor changes in HCC treated by SBRT can affect ADC changes, e.g., cell membrane permeability change, cell apoptosis, necrosis, and fibrosis. But these cellular changes are not specific for the ADC changes. Recently, in vivo microstructural imaging based upon water diffusion MRI has spurred significant interest as a more specific means of modeling factors that influence water diffusion. This is accomplished by acquiring multiple diffusion images with varying known diffusion measuring times, which allows for manipulation of the extent of modulation of water signals by the microstructural environment, thus permitting modeling of tissue microstructural parameters. There are two major modeling approaches^15–21 to such mapping. The first approach assumes that the cell is an impermeable object that confines intracellular and extracellular water molecules in their corresponding compartments^19–21. The restriction effects of varied cellular morphologies on intracellular water diffusion as well as the diffusion gradient pulse influence are considered in this model. This model is valid when the diffusion time is much shorter than the time of water molecule’s residence in the cell. An oscillating gradient spin echo (OGSE) has been used to achieve such short diffusion times. The impermeable assumption model has successfully measured microstructural features (e.g., cellularity) in animal xenografts with histopathological validation²⁰ and demonstrated improved detection of HCC carcinogenesis in rats²². The second approach considers permeable cell membrane restrictions to water molecules using a random walk with barriers methodology (RWBM) to model time-dependent diffusion coefficients D(t)¹⁵. In the short-time limit RWBM, D(t)s measured with different diffusion times can be used to estimate the ratio of volume to surface area of cell membrane (V/S), cell membrane permeability (κ), and unrestricted free diffusivity (D₀).

In a pliot study, we have investigated an imaging protocol to acquire diffusion images at four diffusion times (t_d) with b-values of 50–800 s/mm² on a 3T scanner (MAGNETOM Skyra, Siemens Healthineers, Erlangen, Germany) to estimate cellular parameters in HCCs of 6 patients. To acquire diffusion weighted image series at short diffusion times, we used an OGSE with a sinusoid trapezoid oscillating gradient at 50 and 35 Hz. In addition, a pulse gradient spin echo (PGSE) was used to acquire two diffusion image series at different diffusion times. We used breathing-triggering to acquire 18 slices with an in-plane resolution of 3.5×3.5 mm and 7 mm slice thickness. We scanned the 6 patients before SBRT and 48–96 hours after the 1^st but before the 2^nd treatment fraction of SBRT. Examples of HCC microstructure parameter maps pre-SBRT are shown in Figure 1. The microstructural parameters estimated in GTVs of 6 patients with 8 HCC lesions pre-SBRT and after 1 Fx of SBRT are summarized in Table 3. Conventional ADC maps were calculated by fitting mono-exponential function to diffusion weighted images with b-values from 300–800 s/mm². After 1 treatment, ADCs did not show any significant changes (1.4%±12.8%, p>0.8) but tumor cellular parameters, D₀, (V/S) and k had significant changes (p<0.02). One treatment of SBRT could cause cell death that results in a reduction in the cell membrane surface area present per unit volume and thereby an increase in (V/S), and also could decrease metabolic efflux in cells that reduces D₀ values. We will test these early cellular responses and their predictive values for HCC local control after SBRT in future.

Figure 1.

Color-coded D₀ (top left), V/S (button left), κ (top right) and ADC (PGSE) maps in a HCC tumor volume (pink contour). The grey-scale image is a diffusion weighted image at b=300 s/mm².

Table 3.

Microstructure Parameters from 6 patients with HCC

InGTV	Do (um2/nis)	V/S (um)	κ (um/ms)	ADC (um2/ms)
pre-SBRT	2.77±0.07	2.64+0.12	0.067±0.007	0.78±0.06
after l Fx of SBRT	2.29±0.16	5.01±0.80	0.043±0.005	0.82±0.09
Change (p value)	<0.02	<0.02	<0.012	>0.8

In summary, there are only a few of the investigations of ADC changes in HCC in response to SBRT. The reported results are heterogenous using various diffusion image acquisition techniques and quantitative and qualitive analyses. There are very few data acquired during SBRT, which would be critical for determining if ADC changes can be used as an early response marker. Although an ADC increase post-SBRT is reported, the relationship between this increase and local tumor control in HCC is not clear. Also, there is a lack of a standard on how to assess response or progression in HCC post SBRT. New imaging techniques show great promise in addressing issues of imaging cellular responses in HCC, with the goal of predicting tumor local progression risk early during the course of treatment, permitting individualized treatment of HCC by SBRT.