Improved Theranostic SPECT image reconstruction for radionuclide therapy dosimetry: Part 2, Advanced.

BHTF Insight: Invited Author

Advanced information suitable for health professionals, medical physicians, associated scientists and academics.

Overview

Single Photon Emission Computed Tomography (SPECT) is commonly used to evaluate the radiation dose delivered to target structures and normal organs in radionuclide therapy (RNT). But SPECT imaging is hindered by poor spatial resolution, making it difficult to accurately quantify the radioactivity concentration in small target volumes. In most radionuclide therapy protocols currently used, all patients receive the same standardised injected dose [1], with minimal modification to allow for patient size, age, and extent of disease. Nor is the absorbed dose delivered to tumours routinely quantified. In order for the field move towards more personalised radionuclide therapies the limitations in SPECT quantitative accuracy must first be addressed.

The goal was to develop a reconstruction method that improves the spatial resolution and noise properties of reconstructed SPECT images in a theranostic setting. We term this reconstruction approach:

Single Photon Emission Computed Theranostic Reconstruction (SPECTRE)

The SPECTRE reconstruction approach is novel and is the first reported example using PET images (with their superior noise and spatial resolution) to improve the SPECT reconstructed image. The results obtained from this research show that SPECTRE can provide more accurate image-based dosimetry estimates when compared to conventional reconstruction algorithms, such as ordered subsets expectation maximisation (OSEM).

By comparing the performance of existing clinical methods to the method developed during my research, the aim was to demonstrate:

1. How this approach could improve the quantitative accuracy of SPECT reconstructions and image-based absorbed dose estimates;
   
2. That pre-therapy diagnostic PET imaging can be used more effectively in radionuclide therapy planning; and
   
3. That the method could facilitate personalised patient dosing for optimal therapeutic outcomes.

If one could improve the quantitative accuracy in SPECT imaging of the RNT, it could pave the way for more personalisation in RNT and lead to a better understanding of the radiobiological effects from these therapies. Vendor options offering standardised quantitative software are fast approaching and global implementation of radionuclide therapy procedures is rapidly increasing. Consequently, the demand for accurate quantitative dosimetry is becoming imperative.

Limitations in SPECT imaging of RNT

The partial volume effect (PVE) in SPECT imaging refers to the apparent underestimation of the radioactivity concentration reconstructed in a volume of interest. This is due to the limited resolution of the SPECT system and the finite size of the structures being imaged [2].

In SPECT imaging, the gamma rays emitted by the radiopharmaceutical are detected by a gamma camera, which requires collimation to localize origin of the emitted photon. It is the collimator that largely characterizes the spatial resolution of the system. Due to the limited resolution of the gamma camera and the finite size of the structures being imaged, the activity from a small object (such as a tumor) may spread out over multiple pixels or voxels in the reconstructed image.

The PVE causes an underestimation of tumor activity. It can also result in decreased sensitivity and specificity for detecting small tumours when measuring changes in tumor size over time, and monitoring the patient’s response to therapy.

Partial volume correction (PVC) is important in the context of radionuclide therapy dosimetry. It’s particularly important for performing accurate lesion dosimetry analysis because it allows for more accurate determination of the absorbed radiation dose delivered to tumors.

Various PVC methods, such as iterative reconstruction algorithms, deconvolution algorithms, and other post-reconstruction methods can be used to minimize the impact of the PVE in radionuclide therapy dosimetry. These methods aim to improve the spatial resolution of the reconstructed image and improve the accuracy of lesion dosimetry analysis [3].

This helps optimize treatment planning, assess treatment response, and estimate potential radiation-induced toxicities and ultimately leads to more effective and safer radionuclide therapy treatments. Unfortunately, no single method has earned widespread clinical adoption.

Developing a novel SPECT image reconstruction method

We recently proposed the novel “SPECTRE” reconstruction approach to help improve lesion dosimetry. SPECTRE uses a diagnostic PET image to guide the reconstruction of the SPECT data. SPECTRE aims to improve lesion dosimetry estimates by utilizing a diagnostic PET image to guide the reconstruction of the SPECT data. The SPECTRE approach involves combining information from both a diagnostic PET image and the SPECT data during the image reconstruction process. Here is an overview of the reconstruction methodology:

1. Acquire PET and SPECT data: Both PET and SPECT scans are acquired for the same patient, using the same ligand but with different radiolabels. There’s a positron emitting isotope for diagnostic PET, and a beta + gamma emitting isotope for RNT and SPECT imaging.
   
2. The PET reconstructed image is first co-registered to an initial reconstruction of the SPECT data. The SPECTRE reconstruction involves using an advanced algorithm that incorporates the PET image as a constraint or regularization term. Essentially, the PET image is used as a spatial prior to guide the SPECT reconstruction. It provides supporting information about the location and shape of the lesions to help improve the accuracy of SPECT reconstruction. That’s especially important in areas where SPECT data may be affected by partial volume effect or low count statistics.
   
3. Obtain improved lesion dosimetry: The reconstructed SPECT image, which benefits from the spatial guidance of the PET image, can then be used to perform accurate lesion dosimetry analysis, such as estimating the absorbed radiation dose in the tumors or lesions. The improved spatial resolution and accuracy of the SPECT image reconstruction can lead to:
   • more accurate quantification of activity concentration; and
   • more accurate absorbed dose in the lesions; and
   • can help optimize treatment planning, assess treatment response, and estimate potential radiation-induced toxicities in radionuclide therapy.

Thus, the SPECTRE approach leverages the complementary information provided by both PET and SPECT imaging modalities to improve lesion dosimetry in radionuclide therapy.

By using the high-resolution activity distribution from PET as a spatial guide for SPECT image reconstruction, it aims to overcome some of the limitations of SPECT, such as

• partial volume losses; and
• improve the accuracy of lesion dosimetry estimates;

leading to more effective radionuclide therapies.

Experimental Methods

We demonstrated the SPECTRE reconstruction approach using a dual ⁶⁸Ga/¹⁷⁷Lu IEC phantom study and with a clinical example using ⁶⁴Cu/⁶⁷Cu.

Phantom experiment:

Briefly, a NEMA IEC body phantom was filled with a solution of ¹⁷⁷Lu with a sphere-to-background ratio of 8.5:1. The phantom was scanned on a dual-head SPECT/CT (Siemens Intevo.6) with medium energy collimators.

The same experiment was conducted using ⁶⁸Ga, and the data reconstructed in order to produce the PET image used in the SPECTRE reconstruction approach.

SPECTRE images were reconstructed using the Hybrid Kernelised Expectation Maximisation (HKEM) algorithm. This was implemented from the Software for Tomographic Image Reconstruction (STIR) library. Conventional SPECT reconstructions were generated to facilitate a direct comparison, such as ordered subsets expectation maximisation (OSEM) reconstruction with (OSEM+RM), and without (qSPECT) resolution modelling (rm).

Clinical example with [⁶⁴Cu/⁶⁷Cu]SARTATE:

The SPECTRE reconstruction approach was then applied to a clinical theranostic study using [ 64Cu/67Cu]MeCOSar-Octreotate (“SARTATE”) (Clarity Pharmaceuticals, Sydney, Australia) in a trial of subjects who were being treated for unresectable, multifocal meningioma.

A single dataset was selected to test the SPECTRE reconstruction approach corresponding to:

• the 4 hr imaging time point following injection of ⁶⁴Cu SARTATE for PET imagin; and
• subsequently the 4 hr SPECT imaging time point of the ⁶⁷Cu SARTATE radionuclide therapy.

An initial OSEM+RM reconstruction of the SPECT data was used to co-register the 4 hr ⁶⁴Cu PET image to the matched time-point ⁶⁷Cu SPECT data and used as the guiding image in the SPECTRE reconstruction.

The same conventional SPECT reconstructions (as in the phantom example) were generated for comparison. All SPECT (including SPECTRE) reconstructions were resampled and co-registered to the original 4 hr PET image. VOIs were produced on the PET image and were propagated to the SPECT and SPECTRE reconstructed series where SUV_mean and SUV_max were recorded.

Results

Phantom experiment:

Figure 1. shows transverse slices of the PET and SPECT IEC phantom images with the same window applied to each reconstruction. The OSEM+RM 40 iterations with 12 subsets (40it,12s) reconstruction is labelled as “OSEM+RM” and is the study used in all subsequent analyses. The recovery coefficients RC_mean and RC_max versus sphere diameter for the ⁶⁸Ga PET and ¹⁷⁷Lu SPECT reconstructions are shown in figure 2a and 2b, respectively.

Comparing the image-based recovery of the various reconstructions: the mean recovery coefficient (RC_mean) in the SPECTRE reconstruction is higher than the OSEM+RM equivalent in each of the six spheres.

The qSPECT reconstruction (OSEM reconstruction without resolution modelling) has a RC_mean of 0.62 in the largest sphere and 0.15 in the smallest sphere, compared to SPECTRE which has a RC_mean of 0.82 in the largest sphere and 0.47 in the smallest sphere and demonstrating an improvement in recovery by a factor of 30 % and >300 % respectively.

Looking at global spatial resolution estimates obtained from a matched filter analysis we found:

(i) That the global spatial resolution of the ¹⁷⁷Lu IEC, qSPECT reconstructed image, was 18.2 mm FWHM compared to the estimated SPECTRE resolution of 8.8 mm FWHM and suggests an improvement in resolution by a factor of ~ 2

Improving spatial resolution by a factor of two leads to an improvement in volume resolution of a factor of 8 (i.e., 2³ ).

(ii) The OSEM+RM reconstructed SPECT image was found to have a global resolution of 10 mm – which demonstrates that the application of resolution modelling during the reconstruction does well at improving recovery and resolution. But this method alone is prone to noise amplification, and can produce noise related artifacts causing overestimates of the true radioactivity concentration (as seen in figure 2b).

The results highlight that more sophisticated approaches to partial volume correction are needed in order to facilitate accurate small volume dosimetry estimates in SPECT imaging of radionuclide therapies.

Clinical example with [⁶⁴Cu/⁶⁷Cu]SARTATE:

It is evident in the qSPECT image that lesions 1, 3 and 4 have poor recovery (as would be expected) compared to PET, OSEM+RM and SPECTRE reconstructions.

The SPECTRE reconstruction approach saw an average increase by a factor of approximately 2.7 in the mean radioactivity concentration in small lesions when compared to the standard qSPECT reconstruction.

In figure 3, one particular lesion (L4) in the PET image (which is very faint in the qSPECT reconstruction) shows that the OSEM+RM and SPECTRE reconstructions both saw an increase in SUV_mean by a factor of 3.3.

Important SPECTRE Outcomes

The results demonstrate the utility of the SPECTRE reconstruction approach, showing improved quantitative accuracy and image quality. The phantom and patient examples demonstrate that SPECTRE can provide:

• better spatial resolution;
• higher contrast;
• reduced bias; and
• improved noise characteristics;

compared to the conventional reconstruction methods.

Conclusions

SPECTRE is a novel approach to SPECT image reconstruction and is the first example of using PET images (with similar uptake) to guide the reconstruction of the SPECT data.

The SPECT quantitative accuracy and image quality improves our ability to accurately monitor treatment, with better quantification of the dose delivered to small structures – such as, metastatic nodes and widely distributed cancer target lesions.

SPECTRE image reconstruction shows clear improvements over conventional reconstructions that incorporate resolution modelling. SPECTRE has reduced noise and Gibbs-like artifacts compared to conventional reconstructions.

Further investigation and optimisation of the algorithm parameters is needed and should be investigated for several different theranostic oncological applications that use different radionuclides.

SPECTRE will be useful in the reconstruction of low-count SPECT data. This has been the subject of recent work that will soon be published in a special issue of “Frontiers in Nuclear Medicine”. See: https://www.frontiersin.org/articles/10.3389/fnume.2023.1124283/abstract

Acknowledgements

This article is based on work that appears in EJNMMI Physics:

Marquis, H., Deidda, D., Gillman, A. et al. Theranostic SPECT reconstruction for improved resolution: application to radionuclide therapy dosimetry. EJNMMI Phys 8, 16 (2021). https://doi.org/10.1186/s40658-021-00362-x

The authors would like to thank the Computational Collaborative Project in Synergistic PET-MR Reconstruction (CCP SyneRBI) (www.ccpsynerbi.ac.uk) for enabling a visit to the Institute of Nuclear Medicine (INM) at University College London (UCL). The CCP SyneRBI is funded by the EPSRC (Grant EP/T026693/1). We are also grateful to Clarity Pharmaceuticals (Sydney, Australia) for permission to use the ⁶⁴Cu and ⁶⁷Cu SARTATE image data to test the reconstruction methods developed in this work from a clinical trial they previously conducted (ClinicalTrials.gov Identifier: NCT03936426).

Authors and Affiliations for this work:

Harry Marquis*, Yaser Gholami & Takanori Hioki – Institute of Medical Physics, University of Sydney, Sydney, Australia

Daniel Deidda – National Physical Laboratory, Teddington, UK

Ashley Gillman – Australian e-Health Research Centre, CSIRO, Brisbane, Australia

Kathy Willowson, Enid Eslick & Dale Bailey – Department of Nuclear Medicine, Royal North Shore Hospital, Sydney, Australia

Kris Thielemans – Institute of Nuclear Medicine, University College London, London, UK

*Harry Marquis is now a postdoctoral research scholar at Memorial Sloan Kettering Cancer Center in New York City. His current research looks at developing open source dosimetry software for the nuclear medicine physics community. See: https://mirdsoft.org/

References

1. Strosberg J, El-Haddad G, Wolin E, Hendifar A, Yao J, Chasen B, et al. Phase 3 Trial of 177Lu-Dotatate for Midgut Neuroendocrine Tumors. New England Journal of Medicine. 2017;376:125-35. doi:10.1056/NEJMoa1607427.

2. Marquis H, Willowson KP, Bailey DL. Partial volume effect in SPECT & PET imaging and impact on radionuclide dosimetry estimates. Asia Ocean J Nucl Med Biol. 2023;11(1):44-54. doi: 10.22038/AOJNMB.2022.63827.1448. PMID: 36619190; PMCID: PMC9803618.

3. Gillen, R., Erlandsson, K., Denis-Bacelar, A.M. et al. Towards accurate partial volume correction in ^99mTc oncology SPECT: perturbation for case-specific resolution estimation. EJNMMI Phys 9, 59 (2022). https://doi.org/10.1186/s40658-022-00489-5

————————————————————————

Harry Marquis PhD, 26 April 2023