A review of 3D image-based dosimetry, technical considerations and emerging perspectives in ⁹⁰Y microsphere therapy

Abstract

Yttrium-90 radioembolization (⁹⁰Y-RE) is a well-established therapy for the treatment of hepatocellular carcinoma (HCC) and also of metastatic liver deposits from other malignancies. Nuclear Medicine and Cath Lab diagnostic imaging takes a pivotal role in the success of the treatment, and in order to fully exploit the efficacy of the technique and provide reliable quantitative dosimetry that are related to clinical endpoints in the era of personalized medicine, technical challenges in imaging need to be overcome. In this paper, the extensive literature of current ⁹⁰Y-RE techniques and challenges facing it in terms of quantification and dosimetry are reviewed, with a focus on the current generation of 3D dosimetry techniques. Finally, new emerging techniques are reviewed which seek to overcome these challenges, such as high-resolution imaging, novel surgical procedures and the use of other radiopharmaceuticals for therapy and pre-therapeutic planning.

1. Introduction

The primary form of liver cancer (hepatocellular carcinoma - HCC) is the second largest contributor to cancer mortality in the world, and is the second most common cause of death from cancer worldwide, estimated to be responsible for 746,000 deaths in 2012 (9.1% of the total) [1]. In Europe, 63,500 new cases of liver cancer were diagnosed in 2012 [2]. The prognosis for HCC is poor, with 5-year survival rates (dependent on staging) in England approximately 5.5% [2] and in the United States less than 15% [3].

The liver is also a common metastatic site for tumours in organs drained by the portal vein such as from colorectal, pancreatic and stomach malignancies [4]. HCC and liver metastases represent different tumour types, with more peripheral vascularization in HCC and a higher proportion of smaller lesions in metastases [5]. Similar to primary HCC, surgical resection is usually considered the best curative practice, although only approximately 25% of patients are eligible [6].

The liver possesses unique vascular anatomy due to its dual blood supply. Normal hepatic tissue obtains over 70% of its blood supply from the portal vein, while intrahepatic malignancies derive their blood supply almost exclusively via the hepatic artery [7]. Using this process, intra-arterial techniques such as radioembolisation of radioactive microspheres (RE) can selectively target the liver malignancies. RE typically employs 90-yttrium microspheres (⁹⁰Y-MS - β-emitter, t_½ =64.2 hours, E_β(av)=0.94 MeV, average tissue penetration = 2.5 mm max. range = 1.1 cm) of which there are currently two types commercially available; a glass microsphere (TheraSphere, BTG Inc./MDS Nordion Inc., Ottawa, Canada) and a resin microsphere (SIRTeX Medical Ltd, Sydney, Australia) hereon denoted GMS and RMS respectively. The properties of these MS are detailed in Table 1. Since approval of these devices, a rapid increase in the amount of publications indicating their use has been published (see Figure 1).

Table 1.

Parameters of RMS and GMS for ⁹⁰Y-RE

	SIR-Spheres	Therasphere
Radioisotope	90Y
Isotope location	Attached to surface	Incorporated into glass matrix
β emission (MeV)	2.28 (100%)
γ emission	None
Matrix material	Resin	Glass
Density (g/ml)	1.6 [15]	3.2 [15]
Av Diameter (μm)	32±10 [15]	25±10 [15]
Number of particles (range)	30-60 x106 [16]	3-8 x 106[16]
Bq per sphere	50 [16]	2500 [16]
Embolic effect	Mild-moderate	Mild
Available activity (GBq)	3	3, 5, 7, 10, 15, 20
Shelf life	24 hours	12 days
Endpoint	Target Activity/stasis	Target Dose

Figure 1.

The growth in publications related to ⁹⁰Y-RE as of Jan. 1st 2015. An online publication search (http://www.ncbi.nlm.nih.gov/pubmed/) had the following search conditions: text strings “Microspheres” or “radioembolization” to appear in Title/Abstract and “90” to appear in the Title/Abstract. The final publication list was manually filtered for relevant publications and sorted by date.

Evidence has been mounting regarding the ability of a measurement of the absorbed dose to tumour to help predict the tumour response and patient outcome, although dosimetry rarely finds routine clinical use despite response being a key parameter in the clinic [8]. A recent review by the EANM Dosimetry Committee identified 48 papers (out of 79 surveyed) employing the use of radionuclide dosimetry showing a strong correlation between absorbed dose and the tumour response & toxicity over a wide range of radionuclide treatments (including ⁹⁰Y-RE) [9], and many works have noted the lack of randomized trials comparing dosimetry-based radionuclide therapy to fixed dosing, or dosing per kg body weight [10, 11]. For routine use, there is a requirement of standardized procedures for absorbed dose calculations [12] and improved reliability of radiobiological models used to convert dosimetric data to biologic endpoints [13]. ⁹⁰Y-RE treatment planning should be based on a priori knowledge to predict the total absorbed tumour/normal liver doses (hereon denoted D_T and D_NL respectively) and intended responses before or during the therapeutic intervention, as is common in brachytherapy or external beam radiotherapy (EBRT). The principle disadvantages of patient-specific dosimetry for ⁹⁰Y-RE center around logistical aspects and enabling of resources/time [14]. Quantitative data acquisition procedures are complex and demanding and require significant expertise to ensure integrity of data, reduction of errors and reliability of the methodology.

It is with this in mind that current techniques of ⁹⁰Y-RE are reviewed such as factors affecting accurate quantification, 3D-dosimetry techniques, and emerging techniques that may further help the efficacy of the technique in providing reliable predictive clinical endpoints.

2. Current Techniques

2.1. Procedure

Implantation of ⁹⁰Y-MS is typically a 2-stage process, a ‘mapping’ phase and later (usually 1-2 weeks) an implantation phase. The mapping step is carried out using a microcatheter guided via fluoroscopy through the common hepatic artery. Extrahepatic branches to critical organs are identified and prophylactically embolized in order to limit the potential deposition of MS to the critical organs. Although outside the scope of this review, the reader is referred to excellent reviews related to the angiographic mapping procedure [17].

Once the mapping and embolization steps are complete, ^99mTc-MAA (75 MBq – 150 MBq, hereafter denoted ‘MAA’) is injected through the catheter where the albumin particles are trapped in the first arteriolar–capillary bed encountered. The properties of MAA are summarized in Table 2. The patient is then taken to a gamma camera and a scan performed for the purpose of confirming access to the liver by the delivery system, excluding access to extra-hepatic sites, and defining the extent of shunting to the lungs (described below).

Table 2.

Typical parameters of ^99mTc-MAA for pretherapy assessment.

Radioisotope	99mTc
Isotope location	Attached to surface
β emission	None
γ emission	141 kev (89%)
Matrix material	Aggregated human serum albumin
Density (g/ml)	1.1
Av Diameter (μm)	10-60
Number of particles	0.15 x 106
Bq per sphere	-
Embolic effect	Mild
Available activity (GBq)	Any
Shelf life	Dissociation after 2 hours
Endpoint	Target Activity

There is broad agreement on SPECT-CT being more clinically useful in terms of extrahepatic uptake and more accurate in terms of activity quantification than either SPECT or planar imaging [18–20] with a recent study of 58 patient scans showing rates of detection of extrahepatic MAA was 72%, 79% and 96% for planar, SPECT and SPECT-CT, respectively [21]. Other studies have identified extrahepatic uptake on SPECT-CT not discernable on planar or SPECT only, leading to treatment adjustments such as additional surgical embolization [18, 22]. Other work shows a sensitivity of 100% of using SPECT-CT to detect extrahepatic uptake compared to 41% for non attenuation corrected SPECT and 32% for planar imaging respectively [22]. Recently evaluated also is the discordance between hepatic angiography and MAA SPECT, noting severe discordance in only 3 of 74 cases of patients with HCC [23].

The ⁹⁰Y-MS implantation involves repeating the catheterization procedure on the treatment day. Extra mapping may be performed to examine the patency of previously implanted embolization coils, and further embolizations may be attempted as the time difference between the MAA and MS treatment may allow for development of arteriovenous anastamoses. ⁹⁰Y-MS are then infused through the hepatic arteries under fluoroscopic guidance, with caution paid to any reflux of spheres due to potential stasis of hepatic flow/vascular spasm in the hepatic artery. The activity of MS to be implanted is dependent on factors such as the type of spheres used, properties of the target site (i.e. primary HCC or metastatic tumours) and the level of lung shunt determined from the MAA scan, and can include the desired absorbed dose to the tumour and critical organs whilst avoiding radioembolization induced liver disease (REILD). Typically, the patient is again taken to a gamma camera and Bremsstrahlung imaging performed to localize the MS.

2.2. Activity Planning

2.2.1. Resin Microspheres (RMS)

An empirical method was devised based on the intraoperative calculations from beta probes of early clinical trials [24], providing tables of activity dependent only on liver tumour volume. Tumour involvement that was < 25%, 25%–50%, or > 50% of the total liver volume was treated with 2 GBq, 2.5 GBq, or 3 GBq respectively. A U.S. Consensus panel some years ago recommended discontinuation of this method [25], as the high risk of REILD in small livers of early clinical trials using this method was confirmed in a multicenter study [26].

The body surface area (BSA) method administers activity as a function of the liver tumour volume and the patient size by the empirical equation:

$$A[GBq]=(BSA[m^2]-0.2+\left(\frac{V_{tumor}}{V_{liver}}\right)$$ (1)

where BSA (m²) is calculated as 0.20247 × height (m)^0.725 × weight (kg)^0.425. V_tumour and V_liver represent the volumes of tumour and total liver respectively. A multiplicative factor is recommended concerning whole lobar treatments:

$$A[GBq]=\left(BSA[m^2]-0.2+\frac{V_{tumor}}{V_{lobe}}\right)\times \frac{V_{lobe}}{V_{liver}}$$ (2)

This method is the most common because of its ease of implementation, requiring only measurements of the tumour volume and patient height and weight, and an online tool is available to aid the user in activity calculation¹.

Early clinical trials with RMS noted the presence of radiation pneumonitis (RP) post-treatment [27]. Modifications to injected activity were proposed depending on the level of hepatopulmonary shunting of MS through arteriovenous vessels bypassing the capillary bed to the lungs, as an absorbed dose of 30 Gy to lungs (D_LUNG) has been suggested to cause RP [28]. The manufacturer of SIR-Spheres and recent guidelines recommend less than 25 Gy and preferably less than 20 Gy [29, 30]. The lung shunt fraction is calculated as LSF=lung counts/(liver counts + lung counts), and is a multiplicative modification to the ⁹⁰Y activity determined by Eqn 1 (LS<10% LSF=1; 10-15% LSF=0.8; 15-20% LSF=0.6; >20% no treatment).

Via MIRD calculations, the target organ dose is calculated as the product of the cumulated activity in the organ and the corresponding organ S-value [31], and the total target dose is the summation of all source organ contributions. However the use of organ level S-values inherently assuming uniform activity distribution in the organ, and the use of standardized anatomical models are the major limitations to the technique. Complete reviews of the principles of organ-level MIRD dosimetry relating to ⁹⁰Y-MS are available in the literature [32].

The partition model (PM) incorporates tissue masses and a measurement of the tumour-to-normal tissue (TN) ratio[33]. It requires separation of the organ system into compartments (normal liver, lungs and tumour), and setting prescribed safe radiation doses whereby the maximum administered activity does not exceed these dose limits (80 Gy to normal functioning liver [34], [70 Gy for patients with cirrhosis [34]] and less than 25 Gy to the lungs [27]). The activity required for implantation (A_TOTAL) is calculated based on a limiting lung dose, and also for a limiting D_NL:

$$A_{\mathit{\text{TOTAL}}}(\mathit{\text{lung}})=\frac{D_{\mathit{\text{LUNG}}}\times M_{\mathit{\text{LUNG}}}\times LSF}{49670}$$ (3)

$$A_{\mathit{\text{TOTAL}}}(\mathit{\text{normal liver}})=\frac{D_{NL}\times (M_{\mathit{\text{LIVER}}}+TN\times M_{\mathit{\text{TUMOR}}})}{49670\times (1-LSF)}$$ (4)

where M_LIVER, M_LUNG, and M_T represent the masses of normal liver, lungs and tumour (in g) respectively. The TN ratio is the ratio of activity in the tumour (A_T) and normal liver (A_NL) per unit mass (M_T and M_NL respectively) of the compartment as determined from the MAA scan:

$$TN=\frac{A_T/M_T}{A_{NL}/M_{NL}}$$ (5)

It should be noted that this is an estimate of the TN, and may not always be the same as the true TN.

2.2.2. Glass Microspheres (GMS)

The activity for administration of GMS relates directly to the desired absorbed dose based on a nominal target dose (150 Gy/kg), and assumes a uniform distribution of microspheres throughout the liver. A general equation for determining the activity based on 2-compartmental MIRD macrodosimetry is written as:

$$A_{\mathit{\text{TOTAL}}}[GBq]=\frac{D_{\mathit{\text{TARGET}}}[Gy]\times M_{\mathit{\text{TARGET}}}[kg]}{49.7[Gy·kg/GBq]}$$ (6)

where M_TARGET is the mass of the target (i.e. whole liver or liver lobe, or lungs). There are no adjustments to the ⁹⁰Y activity based on LSF; the treatment proceeds provided D_LUNG <30 Gy (single treatment) and <50 Gy (cumulative treatment). The maximum activity is calculated based on a limiting dose of 30 Gy to the lungs taking into account the LSF:

$$A_{\mathit{\text{MAX}}}=\frac{30\times M_{\mathit{\text{LUNG}}}}{LSF\times 49.7}$$ (7)

From Eqn 6, the liver dose can be defined as:

$$D_{\mathit{\text{total liver}}}=\frac{49.7\times A_{\mathit{\text{TOTAL}}}(1-LSF)\times (1-R)}{M_{\mathit{\text{total liver}}}}$$ (8)

where R represents the percentage of the total activity remaining in the vial after treatment. Thus the absorbed dose is highly dependent on how the liver mass is calculated.

2.3. Differences between RMS and GMS

There are clear differences between GMS and RMS and there is ongoing debate as to whether these differences lead to different efficacy and toxicity profiles [35, 36]. Although it has been postulated that the optimal combination of specific activity and embolic load is somewhere between that of GMS and RMS [37], recent work shows that the overall survival of patients treated with either GMS or RMS across different HCC stages is quite consistent [38]. A multicentre study of hepatic metastatic neuroendocrine tumours showed a statistically significant greater median absorbed dose to liver lobes delivered using GMS (right 117 Gy, left 108 Gy), than using RMS (right 50.8 Gy, left 44.5 Gy), although a similar disease control rate for both (92% GMS and 94% RMS were partial responders or stable disease) after 6 months [39]. More fundamentally, computer simulations have demonstrated the absorbed dose deposited around a point source of resin and glass spheres to be with 1% or each other [40].

Recent studies have shown a relatively large difference in mircoscopic radiobiology between GMS and RMS behaviour, and that the lower number of GMS used for treatment provides a less uniform irradiation, thus allowing a mean whole liver absorbed dose (D_WL) of 120 Gy with no toxicity to the radiosensitive portal triads [41, 42], or 150 Gy for single treatment (268 Gy for repeat administrations) [43]. Their results provide reasoning to the D_NL threshold noted in other clinical studies of 70 Gy using GMS [44] and approximately 40–50 Gy using RMS [13, 45]. It should be noted that although no toxicity was noted with absorbed doses of up to 100 Gy [46], the manufacturer of RMS recommends a D_NL of 80 Gy, and a recent summary suggests a preferable D_NL of 50 Gy [29]. Toxicity of either GMS or RMS is not only dependent on the dose volume factors, but also on the patient population, underlying liver disease, liver function and concurrent therapies [47].

Stasis represents the main reason for stopping the delivery of RMS before full planned activity is given due to the high number of spheres (20 million/GBq), and is not desired in part because of shunting into normal liver causing tumour hypoxia [26]. Due to the lower number of GMS typically encountered (400,000/GBq), stasis and embolic effects have yet to be reported [28]. Therefore because of the mechanics of sphere deposition, a higher injected activity may not always lead to a higher D_T. In a recent editorial, it was noted that the specific activity and number of spheres per GBq should be considered a crucial variable and thus reported in clinical trials [48].

3. Current issues in ⁹⁰Y therapy

3.1. Assaying the Activity

Error in assaying the activity against the manufacturer determined activity creates further errors in absorbed dose calculations. Although this verification is a fundamental requirement on which all quantitative measurements are made, measurement of vial activity of ⁹⁰Y-MS remains an important issue due to lack of standardization in measurements [49, 50]. Due to Bremsstrahlung production, the activity measurements made in a dose calibrator vary with sample geometry, vial placement within the calibrator, vial thickness and solution volume. RMS activity can also vary widely when in a settled state compared to suspended state (distributed homogenously) in the vial. For example, a 20% difference was noted in MS activity dependent on the material of the V-vial compared to shipping vial [50]. One study noticed a 2-16% variance of the MS activity when compared to a reference source between 5 sites participating in a multicenter study employing RMS [26]. Owing to these factors, a total dose delivery error on the order of 20% is noted as possible [49].

There are also no current traceable standards in the UK, or USA for RMS, although there is a NIST traceable calibration for GMS. Correction factors can be employed although this is not trivial process [51]. It is recommended to characterize the local activity measurement standard rather than accepting the calibration value of the calibrator manufacturer [50] although a previous method investigated a spectroscopic assay of the ⁹⁰Y positron decay of resin spheres measured an activity 26% higher than manufacturer indicated [52]. A dedicated work package of the ongoing European project MetroMRT (METROlogy for Molecular RadioTherapy) is aiming to provide traceable dosimetry standards for ⁹⁰Y-RE procedures, and is described further below.

3.2. Limitations of current dosimetry models

Recent work has noted that there is no known association correlating a patient’s BSA with liver volume, tumour volume or radiation sensitivity [53]. Inherent in the BSA method is the assumption of a fixed mean TN liver ratio of 1 for all patients, sacrificing accuracy for simplicity, although patients typically present with a more favourable ratio [19]. A recent retrospective study of 45 patients with liver metastases treated by RMS showed that using Eqn. 1, the administered activity does not correlate with D_WL [54], with a 2.5 fold difference in D_WL over all patients. They did however find a correlation between liver weight and D_WL, noting that patients with large livers are relatively under-dosed and patients with smaller livers are overdosed using this technique.

A recent study compared activity planning and dosimetry in 26 patients with RMS using 4 models (BSA, empirical and PM) showing that maximum differences in injected activities between BSA and PM methods vary from 123%-417% [55]. Although it has been noted that the PM would be the preferred method of ⁹⁰Y-RE for every patient [55, 56], its main drawbacks are the reliance on segmentation of tumour/non-tumour for activity determination and dosimetry purposes, and the assumption of concordance between the MAA (from which the TN ratio is calculated) and MS distribution. This assumed equivalency is a fact much disputed by recent studies, especially in liver metastases [57, 58] and is described further below. In clinical scenarios, the PM and GMS equations are employed mainly in patients with hypervascular, large and numerically limited lesions, such as HCC. In many metastatic cases, clear definition of the tumour boundaries proves difficult due to diffuse metastatic spread and differing vascularity [5], and therefor the PM proves difficult to apply. Recent work shows it cannot be recommended in general for patients with liver metastases [57, 59]. This model also disregards the spatial and temporal variation of the dose, dose rate and radiobiological effects, providing a simplified picture for patient dosimetry [60].

3.3. Determination of lung shunt fraction (LSF)

Dissociation of ^99mTc-MAA into free pertechnetate (^99mTcO₄^-) has been known to change the patient LSF classification and also degrade the image quality[61]. Clinical reporting issues also exist relating to dissociation; in a recent study, uptake in the stomach could not be attributed either to pertechnetate or true MAA uptake [61]. In certain clinics, patients undergoing RMS treatment are pretreated with sodium perchlorate before angiography, which aids in preventing unspecific uptake of MAA in the stomach and thyroid [22, 62].

A recent study has shown that increasing LSF results when scanning patients at later timepoints [62]. At <1 hour after MAA injection a LSF of 7.1% was calculated, whereas at >4 hours, an LSF of 21.6% was determined. The kinetics of MAA due to degradation over time have also largely been ignored [49]. Provided the technique is performed correctly, the LSF estimation is reproducible, even despite different catheter positions when repeated on the same patient [61].

The use of scatter correction routines during planar image acquisition for LSF determination in RMS treatments has been shown to have an effect on LSF, with a potential overestimation of 50% potentially moving patients into different LSF classifications [63]. There is a further reduction in LSF using SPECT whereby volumes of interest are used instead of planar regions. It has also recently been shown that without CT attenuation correction (AC), the LSF can be overestimated by up to 65% on average of 26 patients receiving RMS treatment, and without AC would have resulted in reduced MS activity for 7 patients and excluded 1 from treatment entirely [55].

3.4. Quantitative ⁹⁰Y reconstruction

Bremsstrahlung imaging suffers from poor resolution (approximately 10-15 mm), primarily due to the lack of a pronounced photopeak. The wide range (0 – 2.3 MeV) and continuous nature of the spectrum prohibit the use of energy window based scatter rejection/correction techniques. Other processes that confound corrections include AC (based on a single photon energy), collimator scatter, lead X-rays, septal penetration, camera backscatter and partial energy deposition in the crystal [64].

For gamma cameras, a common calibration method is to use a large phantom filled with ⁹⁰Y [65] or a vial of activity [66]. Images are then reconstructed, and a calibration factor derived from region measurements compared to the known activity concentration.

Recent efforts have shown an improvement in the quantitative accuracy of ⁹⁰Y imaging through the use of Monte Carlo (MC) based modeling of the image degrading factors. Simulations can be performed to calculate energy-dependent scatter kernels and collimator-detector response (CDR) tables which can be applied during image reconstruction [65]. This method showed that activity in a large source such as the liver was estimated with a bias of around −70%, when no compensations were included, whereas with compensations, bias was reduced to -10% to 16%. Similar work along with a model of the decay location and photon emission location incorporated into an OSEM reconstruction, showed increased quantitative accuracy validated on digital (XCAT) and physical phantoms with errors between -1.6% and 11.9% [67]. Other work provides “on-the-fly” MC simulations of scatter and attenuation effects for each patient, and is less dependent on patient geometry [68]. They show improved image contrast and decreased count error compared to non-MC SPECT reconstruction, and also a more accurate mean absorbed dose to a phantom. A recent quantitative SPECT dosimetry guideline recommends the use of MC based corrections for Bremsstrahlung imaging [69] however the routine clinical implementation of such corrections is not easily applied.

In a comparison of collimators for ⁹⁰Y Bremsstrahlung imaging, recent phantom work showed that a medium energy pinhole (MEPH) collimator (or multi-pinholes) and the inclusion of a scatter model provided similar results to TOF (time of flight) PET in terms of contrast recovery and image quality [70].

3.5. Using ^99mTc-MAA as a ⁹⁰Y MS surrogate

Using ^99mTc-MAA biodistribution as a dosimetry predictor for MS implantation is an attractive option whereby the MAA scan can be used to determine the desired activity of ⁹⁰Y-MS to be injected to achieve a prescribed D_T. However, different physical properties, surgical procedures and timing between catheterizations result in MAA being an imperfect surrogate for GMS and RMS [23, 71, 72]. Recent discussions observe that discordance in MAA-microsphere distribution may be attributable to the differing vascularity of HCC and metastases [59], with a suggestion of a more concordant relationship in HCC cases [73].

In terms of size, arterioles feeding liver metastases have a diameter of 30-40 μm, and thus larger MAA particles are likely to embolize outside the metastatic tissue [72]. Thus existing differences in flow between metastatic lesions of variable size are likely to be exaggerated on MAA imaging when compared to MS imaging. In recent work, the visual correlation between co-registered MAA distribution and RMS distribution for 20 patients with hepatic metastases ranged from high to very poor, with a mean Spearman rank correlation of 0.65 [74]. A study of RMS in 39 patients (with both HCC & metastatic lesions) segmented the liver into segments, and showed that for 68% of all segments (n=225), a difference of >10% between MAA and MS activity distribution was found, and that in every procedure at least 1 segment showed a >10% difference [58]. A similar study of 81 patients with a mix of HCC and metastases treated with RMS showed 31 patients had segmental perfusion differences, influenced highly by catheter position [75]. A recent study of 18 HCC patients treated with RMS found that the ratio of MAA to MS uptake in the tumour was 1.4:1, indicating that using MAA imaging for dosimetry calculations can overestimate D_T [76].

Surgical procedures have been identified as a cause of MAA-MS mismatch such as differences in injection rates, microcatheter positioning [75] and arterial flow hemodynamics [23]. A 5-10 mm difference in catheter position between tip and artery opening has been shown to have a major impact on microsphere flow distribution [58, 77]. Fluid dynamic models of MS transport in a hepatic arterial system highlight the influence of the injection time interval, specific daughter vessel targeting, microsphere injection velocity, and vessel morphology [78, 79].

3.6. Microsphere distribution

Fundamentally, MS methodologies are based on the assumption of uniform distribution within tumour/non-tumour compartments. Early investigations into RMS localization examined ex-vivo liver tissue samples to measure the distribution of RMS showing clustering of microspheres around the tumour periphery [80]. Approximating a liver neoplasm as a sphere, 90% of microspheres were found in a 6 mm periphery zone around an inner avascular core (50-70 times more spheres in the periphery than normal liver). Further work assuming each sphere as a point source of radiation, showed a large difference in average D_NL compared to the MIRD method (8.9 Gy vs 80 Gy) [81].

Using a MC dose deposition technique, the heterogeneous nature of dose distribution with ranges of 25 Gy to 3,000 Gy over a 4 cm area for 4 explanted livers was shown [15]. An example of this high dose range is shown in Figure 2. This work showed similar clustering for GMS, noting high doses to tumour/normal tissue boundary, and not observing any difference in the embolic location of GMS or RMS.

Figure 2.

Photomicrograph showing a mid-plane tissue sample with RMS (black dots). White dots represent sphere positions from sections anterior and posterior to the slice shown. Overlaid are the 100 Gy (blue) and the 1000 Gy (red) isodose curves. Reproduced with permission [15]

More recent work with RMS in liver biopsies showed on average 8 to 59 spheres per cluster, with an activity concentration of 1040 and 3050 Bq/mg for 2 patients[82]. Similar levels of clustering were also shown in simulations of GMS in different arterial tree models when the difference in number of spheres is taken into account [41].

4. 3 Dosimetry Methods in ⁹⁰Y-RE

4.1. Direct Monte Carlo

By the introduction of a complete anatomical dataset (usually a CT scan) and a quantitative map of cumulated activity (PET or SPECT images) into a generalised model of radiation transport, an accurate means of calculating patient-specific absorbed dose distributions can be obtained through the use of MC simulations. The technique allows the assessment of heterogenous dose distribution in targets via the use of dose volume histograms (DVH).

Direct MC methods are based on general-purpose radiation transport codes of which there are many: EGS or EGSnrc (electron gamma shower), MCNP (Monte Carlo N-Particle), PENELOPE (PENetration & energy Loss of Positrons and Electrons) and FLUKA (FLUktuierende KAskade). A toolkit more tailored to nuclear medicine imaging called GATE (Geant4 Application for Tomographic Emission) through the GEANT4 toolkit is also freely available and a recent review describes its application to radionuclide dosimetry [83]. These platforms can account for radionuclide emissions and tissue density distributions, leading to extremely accurate dose distributions. Studies carried out comparing MC simulation packages to standard MIRD techniques show a negligible difference in absorbed energies between GATE and MIRD in a phantom over a range of monoenergetic photon energies (>50 keV) [84].

The power of these techniques can also be shown at the microscopic scale; Gulec et al employed a MC simulation (using MCNPX) of a 3-D liver model, including various subunits of the hepatic anatomy [85]. They assessed the impact of the lower number of GMS per GBq (compared to RMS), showing that for both techniques, the liver dose for the compartmental and structural models are similar but that doses to the micro-anatomic structures can be more than 3 times larger than the liver dose for GMS over RMS. A similar MC study showed that the non-uniform trapping produced by GMS transport in a simulated arterial tree may explain their lower toxicity per Gy [41]. They showed a highly asymmetric spread of dose distribution, with 17% of lobules receiving < 40 Gy.

Owing to the effect whereby the necrotic tumour core is surrounded by activity as observed in liver explantations, a tumour necrosis model to correct the activity prescription for therapy evaluated using PENELOPE [86], separates the tumour into a sphere of necrosis surrounded by a uniform shell source of ⁹⁰Y activity. The results show that more beta particles escape the source region and deposit their energies in areas other than the tumour region, and the prescribed activity can be adjusted to take account of this geometrical correction. They further note that the prescribed activity is underestimated by more than 10% by not taking account of this correction.

Recent work has shown new advances in MC calculation algorithms, which accounts for tissue density heterogeneities, and allows a marked reduction in calculation time from 40 hours (10⁸ histories using MCNP6) to 4 minutes using a “collapsed-cone” algorithm [87]. A comparison of mean dose for 6 phantoms and 2 clinical cases (1 ⁹⁰Y microspheres) showed a difference of <1% between the methods.

4.1.1. MC User-end codes

A user-end package called OEDIPE (based on MCNPX) developed at ISRN in France uses the patient’s ⁹⁰Y Bremsstrahlung/PET and CT scans, and has recently applied their code to treatment optimization for hepatic metastases to patients treated with RMS [88]. They calculated the maximum injectable activity for the standard PM dosimetry, and also for an MC calculation using the same dosimetry considerations, as well as an MC calculation using DVH based dose constraints (i.e. only 50% of normal liver receiving a maximum of 30 Gy). Their results show that the maximum injectable activity can be substantially higher when compared to the activity calculation provided standard PM by 27% for the MC calculation using PM dosimetry considerations and by 40% when using DVH-based dose constraints.

Another well-established package called 3D-ID (developed in EGSnrc) has been extended to include radiobiological models derived from the linear quadratic equation, with BED (biologically equivalent dose) calculated for each voxel by estimating the clearance rate in each voxel using the radiobiological parameters α, β and μ², and also an EUD value (effective uniform dose) for a particular user-defined volume. The code has been renamed 3D-RD [89] and recent validations were performed against a voxel S-value MIRD technique in a voxelized liver phantom with less than a 1% difference in absorbed dose [90] and more recently to examine the effects of tissue density over 3 different radionuclide treatments (including ⁹⁰Y-RE) with differences of 0.8% in both D_T and D_NL [91].

4.2. Dose kernel convolution methods

Application of kernels in voxel-based geometries can be found in the computation of “Dose Voxel Kernels (DVK)” or “Voxel S-Values (VSV), which is input data suited for analytical convolution with voxel-based cumulated activity maps (i.e. from SPECT or PET imaging) as defined in MIRD Pamphlet 17 [92]. A VSV is defined as the mean absorbed dose to a target voxel per radioactive decay in a source voxel, both of which are contained in an infinite homogeneous tissue medium. Convolution of the S-value dose distribution (Gy MBq^-1 s^-1) with the cumulated activity (MBq s) leads to a map of absorbed dose (Gy).

A comparison of MC codes (MCNP4C, EGSnrc and GEANT4) used to define the S-value matrix for β-emitters has been shown to affect dose distribution values by only a few percent [93]. MIRD pamphlet 17 originally produced VSVs for 3 voxel sizes and 5 isotopes although more recent work expanded these tables using simulations within the EGSnrc toolkit (and validated with PENELOPE and MCNP4c codes) with more isotopes and a larger range of voxel sizes more suited to modern scanners [94]. These VSV matrices are freely available³.

Recent work has shown that provided VSVs are available for a high resolution voxel size from an MC simulation, VSVs for any lower resolution voxel size can easily be generated [95]. Compared to MC simulation for the larger voxels, this rescaling technique showed differences of 1.5% in scored energy deposition for ⁹⁰Y. A similar approach whereby the fine VSV map was resampled to the same voxel size as the activity map was validated by comparisons to MC simulations (using 3D-RD) in a simulated hepatic tumour, showed a difference in absorbed dose of 0.33% for ⁹⁰Y [90]. A further clinical study by this group performed a comparison of the PM with a user-end code called VoxelDose for 10 HCC patients, showing mean relative differences of 1.5% for D_T and 4.4% for D_NL, noting that the entire process (after tissue segmentations) can be performed in 15 minutes [96].

User-end codes are easy to implement for ⁹⁰Y-RE, requiring only a convolution of the cumulated activity (from a single scan) with a VSV/DVK matrix to produce a 3D map of absorbed dose. A code called VoxelDose was validated in an abdominal phantom using ¹¹¹In with the dosimetry measured by TLDs [97]. They showed a range of errors from 3-62% in activity concentration dependent on the organ, which may be explained by the position of the TLDs. Another code called qDOSE is under development at the Royal Marsden (UK), which offers a specific version of the code aimed at dosimetry for ⁹⁰Y-RE using DVKs developed from EGSnrc environment [98]. Another recently developed user-end code called NUKDOS performs VSV dosimetry, and has been validated against OLINDA and other in-house codes in peptide receptor radionuclide therapy (PRRT) [99].

As S-values are defined in a certain medium, the technique does not account for tissue inhomogeneities. Assumptions can be made that heterogeneities do not induce significant errors with self-dose to the liver due to its size and composition. It has also been noted that if tissue heterogeneities cannot be ignored (i.e. for cross-dose calculation), a low number of histories can be used to generate the cross dose to the target organ using a direct MC methodology, and VSVs for the self dose to the organ, and adding the results [90].

4.3. Local Deposition Method

The Bremsstrahlung self-dose from an organ is very small compared to the beta radiation dose (0.2% of the total) [100]. Thus because of the average range of a β-particle and the low resolution of current clinical SPECT/PET imaging, ⁹⁰Y β-particles emitted within a single voxel range deposit energy locally (within the same voxel). Thus the absorbed dose in each voxel can be determined by multiplying the activity concentration by a constant scalar factor. Recently published conversion factors allow these calculations [101]. As voxel sizes increase the differences between the LDM, VSV and direct MC methods are markedly reduced [102]. The LDM technique provides a fast and efficient method to transform activity concentration to absorbed dose and has been finding increasing clinical use, with a reported case study showing that after an infusion of RMS, a secondary infusion of RMS was used in order to provide a dose boost for the HCC to reach an absorbed dose of 120 Gy [103].

4.4. Dose Response Relationship

Despite a correlation between absorbed doses and treatment response and toxicity [9], all evidence supporting a dose response relationship from dosimetry in ⁹⁰Y-RE is based on either non-controlled prospective or retrospective studies.

A recent study reported on the lack of association between metastatic lesion based response (66 patients treated with RMS) and the overall MAA uptake (classified from 1 -no intratumoural uptake, to 4 -strong uptake) and that the metastatic response cannot be predicted by the degree of perfusion on ^99mTc-MAA scanning while also showing no correlation between tumour response and catheter position [57]. This work shows conflicting results with other studies, and many recent letters have discussed these results in detail [59, 73, 104–107]. An important conclusion of these discussions is that in dosimetry comparisons of ⁹⁰Y-RE, consideration should be given to the type of microsphere, quantification method, tumour type and the method of response assessment. Further noted is that large-scale dosimetry comparison study of MAA data and ⁹⁰Y-MS imaging for the same set of patients is lacking [66].

Other ⁹⁰Y-RE studies treating HCC have shown a significant dose response relationship with predictive MAA imaging. In a study of 36 HCC patients treated with GMS, a strong correlation was noted between absorbed dose and tumour response (EASL criteria) and also with overall survival (OS). They determined a threshold dose of 205 Gy which enabled a response prediction with 91% accuracy and 100% sensitivity [108]. Using this threshold in a cohort of 71 patients, the group detailed 17 patients who underwent treatment intensification to achieve D_T > 205 Gy while maintaining D_NL <120 Gy [109]. They show an OS of 11.5 months and 23.2 months for D_T < 205 Gy and > 205 Gy respectively. Similarly, a recent study of 52 HCC patients treated with GMS showed that a threshold D_T of 500 Gy was successful in predicting objective responses (via EASL criteria) in their cohort [110]. In an effort to counteract partial volume effects, they also excluded lesions < 3 cc (approximately 1.8 cm diameter) because of a 20% underestimation of activity for a 1.8 cm diameter sphere at a contrast ratio of 4:1 in a previous phantom study [47]. A recent study showed a correlation between D_T >100 Gy and survival in 18 HCC patients treated with GMS [76], as shown in Figure. 3. Treating 25 patients with liver metastases with RMS and using a dual tracer (both pre-treatment ^99mTc) partition model technique, a correlation between D_T and OS with a threshold dose of 55 Gy (32.8 months vs 7.2 months) was noted [111].

Figure 3.

Example of overall survival curves with a tumour threshold of 100 Gy using GMS dosimetry. Reprinted with permission [76].

MAA distributions of 39 liver lesions and an LDM technique was used to show correlation between D_T and metabolic response by FDG PET [112]. Similarly, a mean D_T for 30 tumours using Bremsstrahlung imaging and ⁹⁰Y DPKs showed that the percentage of tumour volume receiving >50 Gy significantly predicted a decrease in SUVmax (assessed with FDG PET), and maximum D_T predicted a decrease in tumour metabolic activity.

Although there has been much work on the radiobiological aspects of ⁹⁰Y-RE, convergence on a consensus approach to radiobiological model is required [113]. Currently dosimetry calculations assume a single set of radiobiological parameter values for cells making up a tumour or organ, although radiosensitivity of tissue components is known to be spatially dependent [114]. A VSV method in 13 patients with liver metastases treated with RMS attempted to incorporate the EUD in place of D_T, and noted that α should be redefined for both normal liver and specific tumours before radiobiological quantities can be successfully employed [115]. Only recently has it been recommended that in any liver dosimetry analysis BED should be considered along with absorbed dose [47].

In a study of 73 patients with HCC, a TCP-NTCP radiobiological model and patient dosimetry calculated by DVK showed that the model predicted TCPs (complete or partial response) of 73% and 55% according to EASL and RECIST criteria using a mean D_T of 110 Gy [13]. They also observed (and predicted) toxicity of 34%, and calculated a BED50 of 93 Gy, close to the values of 72 Gy derived from EBRT [116]. Incorporating the microscale distribution of dose in a single hepatic lobule, the model was adapted and provided a method for computing NTCP as a function of microsphere activity and targeted liver volume [42].

A recent method to assess D_NL involves an injection of ^99mTc-sulfur colloid post ^99mTc-MAA SPECT, which allowed a determination a threshold dose of 24.5 Gy to predict biochemical toxicity [111]. This method allows a more accurate assessment of the functional liver rather than relying on morphologic partition modeling, and further investigations are certainly warranted.

5. Emerging Directions

5.1. ⁹⁰Y PET Imaging

Although many groups have preliminarily investigated ⁹⁰Y PET imaging in terms of post-treatment imaging, it remains an underutilised modality of post-therapy imaging. Adequate visualisation can be achieved with a 20 minute per bed scan [117, 118] and sensitivities of 0.11 cps/MBq in patients [119], 0.577 and 1 cps/MBq in phantom [117, 120] have been reported. In a recent study, the optimal contrast for hot and cold spheres in a phantom was obtained with LYSO crystals and TOF technology [121]. Echoing this statement, a minimum detectable activity of 1 MBq/ml in a phantom was noted on a Siemens Biograph 40 for TOF reconstruction (3 MBq/ml in non-TOF) [122]. They also show the effect of PET image reconstruction on the activity recovered in a phantom, and that differences in activity on the order of ±20% can be obtained depending on the type of reconstruction performed.

High random to true ratios during imaging have shown that a positive mean bias of 3.3±6.4 Gy (averaged over 65 PET scans) can exist due to delayed random coincidence correction producing negative sinogram values, and these negative values being truncated prior to image reconstruction [123]. A physical and mathematical phantom with filled spheres and convoluted via a VSV technique showed approximately 9% difference (using PSF, recovery coefficients and TOF) in absorbed dose when compared to a known ground truth [124]. They noted that the use of the TOF in the reconstruction improved the accuracy of absorbed dose calculations for the simulated tumours with the lowest activities.

Many studies have shown an enhanced resolution of microsphere distribution around the tumour periphery with ⁹⁰Y-PET and also show potential to identify areas of sub-optimal microsphere deposition not detectable with gamma camera imaging [20, 117], as shown in Figure 4. In a recent diagnostic image reporting study of target to non-target activity, ⁹⁰Y PET consistently outperformed ⁹⁰Y Bremsstrahlung [118].

Figure 4.

Comparison between ⁹⁰Y-PET (c) and ⁹⁰Y-Bremsstrahlung images (d) showing the superior resolution of PET in terms of highlighting further small tumours (red arrows) clearly observed in CT (a and b). Also observed is the necrotic core of the largest tumour towards the left of the image. Reproduced with permission [118]

Although much of the recent work has involved calibration of PET scanners to investigate quantitative accuracy, studies have also used PET to provide retrospective dosimetry. HCC lesion dosimetry using rescaled VSVs (using MCNP) showed a mean D_T of 139.3 Gy and D_NL of 33.8 Gy [120] and this work noted a dose of 287 Gy to the tumour periphery, and 70 Gy to the necrotic core.

Absorbed doses to 23 HCC patients with RMS using a LDM and ⁹⁰Y PET, used DVHs to account for the heterogenous nature of microsphere distribution to determine a threshold of D₇₀>100Gy for complete response of HCC [66]. Providing the first Gy-for-Gy comparison against MAA SPECT dosimetry for 7 patients, they showed an intraclass correlation of 0.97, in the cohort, with a maximum difference of up to 40% in D_T. Similar work showed that in 64 patients with HCC treated with GMS and using VSVs that D_NL was 93±33 Gy with on average V₅₀ > 79 Gy and D_T of 173±103 Gy [125]. Recent work used FLUKA to develop a DVK technique for PET-CT scans for 6 patients (3 GMS, 3 RMS) and used DVHs to show D₇₀ in the range of 25-155 Gy for all patients, and a mean D_T of 71-311 Gy [126]. VSVs and PET were also used for 5 patients treated with RMS showing a range of D_T from 51-163 Gy. This work also calculated the BED and converted this to dose equivalent to delivery at 2 Gy/fraction, upon which limits for TCP and NTCP are published. They further showed that the differences in D_T using the PM from the MAA scan and those calculated from VSV PET-CT were in the range of -53.8% to +178.4%. The emerging role of quantitative PET-CT in ⁹⁰Y-RE is further developed in a recent review [127].

5.2. Preplanning with PET isotopes

Efforts have been made to supplant ^99mTc-MAA imaging with microspheres labeled with a positron-emitting isotope, which would provide higher resolution imaging and enable improvements in PET predictive dosimetry.

Recent work has labeled RMS with ⁶⁴Cu, ⁸⁶Y and ⁸⁹Zr with efficiencies of >95% (although 80% for ⁸⁹Zr)[128]. ⁶⁴Cu was excluded due to leaching of Cu(II), and ⁸⁶Y as it emits photons detected in the PET imaging window. They showed good ⁸⁹Zr imaging characteristics in phantoms and a lower cost associated with production [129]. Given the comparable half-life of ⁸⁹Zr to ⁹⁰Y, the group then produced ¹⁸F-based RMS [130]. They achieved a labeling efficiency of 95% and an in vitro stability study of >99%, showing histological analysis of microsphere distribution in the rabbit liver agreed with PET findings to within 3%. They showed however, that 15% leaching occurred at 45 min post-injection and thus it would prove difficult to provide accurate activity quantification using ¹⁸F microspheres. A recent review highlighted the issues inherent with “non-pure” positron emitters such as ⁸⁶Y, ¹²⁴I, ^94mTc because of prompt gamma rays emitted in the PET acquisition window, and also higher energy gamma rays that may scatter into the same window [131]. Significant corrections are required in order to accurately image these isotopes especially for quantitative 3D PET where the potential for spurious coincidence contamination is much increased. Other efforts into using biodegradable Chitosan glycol MS labeled to ⁶⁸Ga have been developed in preclinical models [132], although no clinical results are yet available.

5.3. Other Treatment Isotopes

The Utrecht group have successfully produced poly (L-lactic acid) microspheres loaded with holmium-166 (¹⁶⁶Ho-PLLA)[133], and proposed a phase 1 trial (called HEPAR) to treat patients with liver metastases in 2010 [134], with a dose-escalation protocol of 20, 40, 60 and 80 Gy using MIRD macrodosimetry (achieved with varying GBq/kg of liver weight). The properties of the microspheres are shown in Table 3. A gamma emission at 81 keV enables SPECT imaging and the spheres are biodegradable, meaning no permanent embolization occurs. GATE simulations of a liver tumour and gamma camera details a better localization potential of ¹⁶⁶Ho SPECT than ⁹⁰Y Bremsstrahlung images [135].

Table 3.

Properties of ¹⁶⁶Ho-PLLA microspheres [134].

Name	166Ho-PLLA-MS (UMC Utrecht)
Radioisotope	166Ho
Isotope location	Incorporated into matrix
β emission (MeV)	1.77 (49%), 1.85 (50%)
γ emission (keV)	80.6 (6.7%)
Matrix material	PLLA
Density (g/ml)	1.4
Av Diameter	30±5
Number of particles	33×106
Bq per sphere	≤450
Embolic effect	Mild-moderate
Available Activity	Any
Endpoint	Target Dose

In a porcine model, the group showed good correlation of biodistribution of scout doses (250 MBq) and treatment doses of ¹⁶⁶Ho-PLLA [136]. The results of the HEPAR trial (15 patients) [137], show the maximum tolerated radiation dose to the whole liver for the cohort to be 60 Gy (toxicity occurred in 2 of 3 patients receiving 80 Gy), and proposed this as an endpoint for a phase 2 trial.

As ¹⁶⁶Ho-PLLA microspheres are paramagnetic, they can also be observed on MR imaging via the microsphere induced change in relaxation rates per mg/ml from a baseline scan acquired before therapy [137]. Activity quantitation was achieved using the specific activity of the spheres, and a good correlation (R²=0.91) was shown between MR and SPECT calculated dosimetry to liver segments determined from a convolution with a DVK [138]. From their work, administration under real-time MR imaging may be a technical possibility.

¹⁴²Pr has also been identified as another possible choice for RE (t_½ =19.12 hrs, E_βmax= 2.16 MeV, E_□=1.57 MeV), and a simulation study compared dose distributions of ¹⁴²Pr and ⁹⁰Y within a hepatic tumour and blood vessels using MCNPX [139]. They showed a higher BED of 301 Gy for ¹⁴²Pr against 195 Gy for ⁹⁰Y. Although this represents very early work, it may be an emerging isotope for future trials.

5.4. Delivery of Spheres

A recent study has shown that D_NL can be reduced by the use of temporary degradable starch microspheres (DSM), which lead to temporary redistribution of blood flow away from the embolised tissue [77]. The technique was applied in cases where catheter repositioning was not possible due to location of metastatic deposits for 5 patients and MAA SPECT before and after DSM application showed further selectivity of the tumour, while sparing normal liver tissue. There were no adverse events or evidence of REILD on follow-up.

A recent technical development in the catheterization system (called the Surefire Infusion System) provides a way to bypass coil embolisation. During the RE procedure, retrograde passage of embolic particles is prohibited by an expandable tip, which opens under retrograde flow conditions and collapses under forward flow. This technique has so far been performed on small cohorts of patients, and allowing treatment on a patient who would otherwise not have been treated with the conventional microcatheter technique [140, 141]. The catheter results in a reduction in procedure time, and may represent the next step in standard RE treatment.

5.5. “Extended shelf-life” Technique

A method to balance the desired embolic load with microsphere activity can be achieved by taking advantage of the long shelf-life of GMS (12 days) [48]. The “extended shelf-life” technique administered an average increase of 111% in embolic load (of lower specific activity 266±38 Bq per sphere) to a cohort of 50 patients, almost doubling the required number of GMS to achieve 100 Gy to normal tissue [37]. Their results show better tumour coverage (perhaps due to an increased number of clusters of spheres) and EASL response rates in patients with extensive tumour burden or marked tumour hypervascularity. In a follow-up study of 134 patients, they showed similar response rates with the addition of 84 patients, and also validating the safety and efficacy of the technique [142].

6. Conclusions

The shift towards personalised radionuclide therapy is an inevitable trend, however dose-response effects in terms of toxicity to normal organs and survival must first be proven as part of a randomised clinical trial. Noting that techniques in ⁹⁰Y-RE dosimetry are well advanced, but confidence in the techniques needs to increase, Kao states “the major barrier to the routine application of predictive dosimetry for ⁹⁰Y-RE is no longer the state of the art but rather the state of our hearts” [143] indicating that while many of the techniques for predictive and retrospective dosimetry are already available, the effort involved in employing them, the reliability of the results provided by them, and more crucially, trusting their accuracy and precision, is not a trivial undertaking. This lack of confidence potentially stems from the current limitations of the therapeutic procedure, ranging from MAA-MS imaging mismatches, poor imaging resolution, quantification errors and also fundamental errors associated with MS activity measurements in ion chambers.

As mentioned in this review, studies performing dose escalation based on partition model, MIRD techniques or BSA dosimetry are making important clinical decisions based on dosimetry results, showing belief in the methodology. Thus great attention to detail should be placed on the radiobiological considerations and fundamental quantification methodology to allow accurate absorbed dose determination. The spatial and temporal variation of dose rate will have varying radiobiological effects on tumour control and normal liver sparing and due to the dose gradient, each voxel of the tumour and healthy liver receives a different absorbed dose at a different time. Current efforts into modeling at the microscopic scale will hopefully help address these issues. Noting that many published papers fail to describe their dosimetry methodology, the EANM Dosimetry group have published guidelines and a documentation checklist of all the relevant parameters for reporting of the parameters and procedures involved [144], which should help to enable results and techniques to be compared cross-site. This standardisation becomes important as ⁹⁰Y-RE begins to be applied to tumours in other body sites such as lung metastases through the bronchial artery [145].

The ongoing MetroMRT⁴ project aims to address some of the fundamental issues highlighted in this review, such as standardisation of microsphere calibration in an ion chamber and standardised quantification procedures. Specific work packages were created to examine procedures of ⁹⁰Y-RE [146], and the expected outcomes of the project will be recommendations for standard practice. The French project DosiTest⁵ aims to evaluate the impact of the various steps that contribute to the realization of a dosimetric study by the use of a virtual multi-center intercomparison based on Monte-Carlo modeling. They employ scintigraphic images generated at a core lab using GATE, with users analyzing them with their own dosimetry protocol [147]. The aim is to evaluate bias of certain dosimetry approaches, and to propose a reference methodology applicable in a clinical setting.

The use of patient-specific Monte Carlo techniques has increased rapidly as the computational requirements are more easily met. Platforms such as GATE, EGS and MCNP are being increasingly employed for their versatility for using patient specific scans to produce reference absorbed dose calculations. A typical patient-specific calculation can be performed using GATE in 16 hours (with 2% uncertainty), within the conventional pretherapeutic imaging to therapy time [83]. As these platforms and applications advance, further user end codes are expected to be available which may help to remove the barrier of detailed technical knowledge of the programs and concentrated resources required to perform these simulations in a clinical setting.

The emerging techniques described in this work are aiming to address certain deficiencies in the current technique, such as ⁹⁰Y-PET, which can provide high resolution imaging and does not require considerable corrections to the imaging process in order to provide quantitative values. Ideally this would be preceded by a PET pre-therapy scan of similar resolution for dosimetry purposes, which is not yet clinically available. Other efforts such as the use of ¹⁶⁶Ho-PLLA make SPECT imaging more reliable and more easily quantifiable due to the gamma ray emission, and its use in pre-therapy imaging ensures that the imaging discordance as a result of particulate differences should not be an issue.

Reducing the time the patient spends in the clinic is also an important factor for advancement of the therapy. A recent proof of concept study evaluated the potential for performing cone-beam CT, angiography with pretherapeutic embolisations, planar MAA imaging, LSF determination, dosimetry calculations (from MAA images) and GMS implantation as a single-session treatment as an outpatient procedure [148]. 14 patients underwent the modified procedure, and the average procedure time was 2.7 ± 0.72 hours (maximum of 4 hours). Although no post-therapy imaging is performed, it shows great potential for future applications in terms of reduced time, cost and resources but also requiring an increase in coordination between the relevant multidisciplinary teams.