Patient characteristics
Patients presenting with HCC, CRC, NET, melanoma, and other metastatic lesions enrolled in 90Y treatment at Michigan Medicine from 2016 to 2019 were examined for this study. A total of 35 patients were treated covering 120 lesions with 90Y glass microsphere intraarterial injections. These injections ranged from 0.5 to 12.6 GBq per treatment. Patients receiving treatment to both the right and left lobes were considered independent cases, for a total of 39 treatments. Patients included in this study had a minimum of 1 and a maximum of 9 tumors each that ranged in size from 2 to 871 ml. Subjects with HCC, all cirrhotic, comprised 38% of cases and 36% of all lesions. Among metastatic diseases, colorectal, adrenal, and cholangiocarcinoma lesions were the most commonly included with 4 patients each. Melanomas and neuroendocrine liver metastasis were treated in 3 patients, and pancreatic and appendiceal metastasis had one patient in this study apiece. All patients were Child-Pugh class A at baseline prior to treatment.
Imaging
All patients underwent pretreatment 99mTc-MAA planar and SPECT/CT imaging on Siemens Symbia systems (Intevo or T series) for clinical purposes to estimate lung shunt and extra-hepatic deposition. The acquisition parameters were 15% photopeak and an adjacent 15% scatter window; 128 × 128 matrix, 4.8 mm pixels, 60 views/head; non-circular orbit; and 10–20 min acquisition time. The images were reconstructed using 8 iterations, 4 subsets of Siemens 3D-OS-EM software (Flash3D, Siemens Medical Solutions, Malvern, PA) including CT-attenuation correction, triple-energy window scatter correction, resolution recovery, and a 8.4-mm Gaussian post-filter. The voxel size was 4.8mm3. The CT was performed in low dose mode (130 kVp; 80 mAs) during free-breathing.
Patients in the study signed an informed consent document to participate in the research 90Y imaging. Post-treatment 90Y imaging within 4 h after microsphere injection was performed via a Siemens Biograph mCT PET/CT with time-of-flight (TOF resolution 530 ps). Patient PET data were reconstructed with Siemens 3D-OSEM software using the following parameters that were chosen based on a previous [9] 90Y phantom evaluation: TOF, 1 iteration 21 subsets, attenuation correction, scatter correction, randoms correction, resolution recovery, and a 5-mm Gaussian post-filter. The PET matrix size was 200 × 200 with a pixel size of 4.07 × 4.07 mm2, and slice thickness of 3 mm. The CT was performed in low dose mode (120 kVp; 80 mAs) during free-breathing.
The perfused liver volume was defined via pre-treatment angiography and approved by an interventional radiologist. Each perfused liver volume is composed of the liver segments or lobe being supplied by the artery to be infused [10]. Tumor volumes > 2 mL were segmented manually on baseline diagnostic CT or MR by an experienced radiologist specializing in hepatic malignancies. The diagnostic CT or MR was then rigidly registered to the co-registered 99mTc-MAA SPECT/CT and 90Y SPECT/CT and contours were transformed within MIM version 6.9 (MIM Software Inc., Cleveland, OH). Fine manual adjustments were made as needed to account for mis-registration. The liver was segmented directly on the CT of SPECT/CT and PET/CT using deep learning or atlas-based semi-automatic tools within MIM. The healthy liver was defined as the liver minus the contoured lesions.
In order to correct for partial volume effects that are a consequence of finite spatial resolution, volume-dependent recovery coefficients (RCs) were applied to mean counts in lesions and non-tumoral liver for both SPECT and PET. The RCs for 90Y were measured as part of a previous study [9] and were repeated for Tc-99 m using the same phantom setup (6 spheres of volume 2 – 113 mL with a sphere-to-background ratio of 9:1). The fitted RC vs. volume curves from these measurements are shown in Fig. 1.
Dosimetry via the single compartment model
The current standard for dosimetry in 90Y SIRT is based on the Medical Internal Radiation Dosimetry (MIRD) formalism, which provides an estimate of the average energy deposition per GBq (\( {\overline{E}}_D \)) by utilizing an average beta energy absorption per nuclear transition (\( {\overline{E}}_{\beta } \)) as a substitute for a full transport simulation such as Monte Carlo [11]. This allows for a quick calculation of absorbed dose to the volumes of interest and is appropriate for non-penetrating relatively low energy beta emissions. With an average beta energy of 933 keV per disintegration emitted in the decay of 90Y, the majority of the dose is deposited locally around the microspheres. This high percentage of locally deposited energy keeps the non-penetrating MIRD model accurate and is used for all absorbed dose calculations in this study. Under the assumption that the microspheres are permanently lodged inside the liver with no biological clearance and a physical half-life (T1/2) of 2.66 days the total energy deposited in a volume of interest (VOI) can be calculated and scaled based on injected activity.
When all of these factors are accounted for the average energy deposited by 90Y is 49.6 ≈50 J/GBq. Assuming non-penetrating radiation and uniform activity for any VOI (e.g., tumor, non-tumoral liver, or perfused lobe), the absorbed dose (DVOI) to the VOI per unit activity injected (A) to that volume of mass (MVOI) containing activity (AVOI) is given by [12]:
$$ \frac{D_{VOI}}{A}\left(\frac{Gy}{GBq}\right)=50\ \left(\frac{J}{GBq}\right)\cdotp \frac{\frac{A_{VOI}}{A\kern0.5em }}{M_{VOI}\ (kg)\ } $$
(1)
All delivered tumor and normal liver absorbed doses were calculated according to equation 1 using the defined contours, an assumption of density = 1.03 g/cc, and the direct 90Y PET/CT image-based activities. Under the assumption that all injected activity is in the infused liver lobe and lung, the injected activity, A, required to deliver a dose D to the lobe can be expressed as:
$$ {A}_{Lobe}=\frac{D_{Lobe}(Gy)\ast {M}_{Lobe}(kg)}{50\ \left(J/ GBq\right)} $$
(2)
Multi-compartmental (partition) model
As with the standard model, the partition model uses the MIRD model’s average energy deposition of 50 J/GBq as well as the injected activity scaled by the lung shunt fraction for dose calculations. Instead of a singular lobar or segmental mass, the partition model requires the mass of the normal liver (MNL) and the cumulative mass of all defined lesions (MT). The tumor to normal liver uptake ratio (TNR) can be measured by the tumor uptake of MAA relative to the normal liver parenchyma. Planning or predictive dosimetry uses this ratio to distribute the activity between the mass compartments. In this work, TNR was determined via 99mTc-MAA pretreatment SPECT scans for the radiologist defined tumor and non-tumoral liver. Selecting an absorbed dose for normal liver D given the masses of the normal liver and tumor compartments and lung shunt fraction, an injected activity A, can be calculated according to equation 3. For an injected activity A, these can be used to calculate the absorbed dose per unit activity for either the normal liver, as seen in equation 4, or to the tumor, equation 5 [12].
$$ A=\frac{D_{NL}(Gy)\ast \left({M}_{NL}(kg)+ TNR\ast {M}_T(kg)\right)\ }{50\ \left(J/ GBq\right)\ast \left(1- LSF\right)} $$
(3)
$$ \kern0.75em \frac{D_{NL}}{A}\left( Gy\ per\ GBq\right)=50\ \left(J/ GBq\right)\cdotp \frac{\left(A\cdotp \left(1- LSF\right)\right)/A}{M_{NL}\ (kg)+ TNR\cdotp {M}_T\ (kg)} $$
(4)
$$ \kern0.5em \frac{D_T}{A}\left( Gy\ per\ GBq\right)=50\left(J/ GBq\right)\cdot \frac{\frac{A\cdot \left(1- LSF\right)}{A}\cdot TNR}{M_{NL}(kg)+ TNR\cdot {M}_T\ (kg)} $$
(5)
Absorbed dose parameterization for TCP comparison
In order to observe changes in TCP corresponding to prescribing 90Y using the standard model versus the partition model, a parameterized chart of absorbed dose ranges was created for the retrospective analysis of 120 lesions over 35 patients. The TCP curve (Fig. 2) used in this work was from our prior report [9] where a logit model was used to fit mean tumor absorbed dose-response data in patients who underwent 90Y PET/CT imaging following SIRT at our institution.
As per the package insert [8], the standard model prescription absorbed dose ranges were allowed to vary between 80 and 150 Gy to the infused liver volume depending on tumor staging, cirrhosis status, and estimated percentage of the microspheres that will end up shunted to the lung (LSF), as well as the estimated nominal residual activity in vial and tubing. Partition model prescription ranges were chosen by consulting the TheraSphere Global Dosimetry Steering Committee guidelines of 75 Gy to the normal liver for lobar treatments [1] as well as NTCP curve data showing close to 100% normal tissue complications being reached at 150 Gy to the normal liver [5]. These numbers served as a rough guideline for choosing the 40–150 Gy absorbed dose range of prescriptions presented in this paper.
For each of the prescribed dose values to the perfused liver volume (for SM) and normal liver (for PM), the hypothetical injected activity A was calculated using equations 2 and 3. The corresponding absorbed dose to individual tumors (not the entire compartment) was calculated by scaling the previously calculated absorbed dose per injected activity from the post-therapy 90Y PET/CT by this hypothetical injected activity using equation 1. Then, using our previously generated curve of Fig. 2, the TCP corresponding to each tumor absorbed dose value was calculated for the SM and PM and the change in TCP between the two prescription models was determined. This process was repeated for every combination of SM and PM prescriptions to generate a parameterization chart.
NTCP Comparison
The predicted normal tissue outcomes were calculated via the new radiobiologic model, equation 8, put forth by Walrand et al. [13, 14]. For the current study, the whole liver absorbed dose that gives NTCP = 0.15 (WLTD15) was calculated using equations 6, 7, and 8 and was compared with the whole liver dose associated with a standard model 120 Gy absorbed dose prescription to the infused volume. This model uses external beam radiotherapy treatment (EBRT) clinical data with an endpoint of radiation-induced liver disease. The WLTD was calculated assuming a NTCP (p) for microsphere treatments according to the volume fraction of the liver that is targeted (Vf), the killed lobule fraction (Kf), and the specific activity of the microspheres at the time of injection (msA):
$$ Kf(p)=0.4\ast \sqrt[8.29]{\frac{p}{1-p}} $$
(6)
$$ \kern1.75em F(msA)=\left(1-{e}^{-\sqrt[3]{\frac{msA}{0.0471 kBq}}}\right) $$
(7)
$$ WLTD\left(p, Vf,\mathrm{m} sA\right)=47.1\ Gy\ast \frac{\left(1+0.457\ p\right)\ast F(msA)}{{\left( Vf- Kf(p)\right)}^{0.869\ast F(msA)}}\ast Vf $$
(8)
where F is a dimensionless scale factor that is a function of msA and is related to the average inter-microsphere distance [14].
The NTCP model is based on previous work from external beam therapy showing normal tissue complications to be near zero as long as the portion of treated liver was under 25% to 40% of the total liver volume [15]. Patients with segmentectomies or smaller lobular targets were therefore not included in the NTCP prediction data, as normal tissue complications are not expected to impact patient care in these cases due to the nature of the liver as a parallel organ and small treatment area. In this study, only 29 of the 39 treatments had targeted liver volume fractions large enough (> 40% of the total liver) for the calculation of NTCP.