Doubling PET Imaging Capacity with a New Reconstruction Algorithm

Positron Emission Tomography has enabled quantitative, non-invasive imaging, yet this imaging modality has traditionally seen only one radiotracer at a time. Here we show how multiplexed PET, a new reconstruction algorithm can see two radiotracers simultaneously without any prior tracer knowledge.
Doubling PET Imaging Capacity with a New Reconstruction Algorithm

Positron Emission Tomography (PET) is a non-invasive imaging modality that has made huge advances in the diagnosis of cancer, as well as in the fields of neurology and infection. The power of the imaging modality relies on tethering a radioactive isotope, known as a positron emitter with a molecular imaging agent. Isotopes are paired with the molecular imaging agent based on the circulation time, though one element has remained constant for most of PET; only one isotope should be used for imaging as the reconstruction method only “sees” 511 keV photons after positron annihilation. Despite this limitation, PET is incredibly useful when powered with imaging agents to specific antigens or targets, allowing tissue to be identified as positive or negative for the marker, and intensity can be quantified. This goal toward pure 511 keV emission led PET isotope development to focus on properties that provided the best resolution and abundance while limiting noise.


This project to enable simultaneous dual-isotope PET imaging started in 2011, within the Madrid-MIT m+Vision Consortium, an innovative and bold program funded by the Madrid regional government and directed by Marth Gray from MIT, that pursued the development of innovative and impactful medical imaging projects to be performed by a selection of postdoctoral fellows from all over the world in a collaboration with research groups in both Madrid and Boston. One of the medical needs identified within this program was the question of how to improve the imaging density in PET enabling multiplexed PET acquisitions for a variety of preclinical and clinical applications. Multiple radiotracer PET had already been proved using compartment modeling or prior knowledge of the radiotracer distribution. That same year, A. Andreyev, A. Celler, had shown with simulations that it could be possible to perform a dual-isotope PET acquisition combining positron-emitter and positron-gamma emitter isotopes.  NOTE: Prompt-gamma emitters such as 86Y, 124I, and 52Mn  are PET isotopes that emit one or more additional photons together with the positron. These non-standard PET isotopes are considered “dirty”, as the additional photons typically introduce a significant background to the PET images in standard acquisitions. However, these additional photons may be used to differentiate with respect to standard PET isotopes such as 11C, 18F, 64Cu, or 68Ga.


Therefore, within the m+Vision Consortium, Joaquin L. Herraiz, started working with Vicente Parot, Shivang R. Dave, and Eduardo Lage, to further develop this approach and perform dual-isotope in-vivo studies using real PET scanners in Madrid and Boston. In that initial work, A. Andreyev and A. Celler assumed that the additional gamma rays could be simply identified by their energy. However, most PET scanners operate with a fixed energy window, and the energy information is not usually available. Within the m+Vision Consortium, the acquisition and reconstruction software of the preclinical Argus PET/CT scanner (from the Spanish company Sedecal Molecular Imaging) was adapted to enable this energy differentiation and perform dual-isotope PET imaging. Despite the encouraging initial results obtained with Stephen C. Moore and Mi-Ae Park at BWH in Boston, and Juanjo Vaquero and Francisca Mulero in Madrid, the performance of that initial developed method was limited, as several data corrections (such as randoms, scatter and sensitivity of triple coincidences considering the attenuation of the additional photons) were not accurate enough. Furthermore, as that approach was developed for one specific preclinical scanner, it was not easy to translate to other PET scanners, especially clinical ones, where modifications of the firmware is not allowed.


Right when these limitations seemed hard to overcome, Joaquin L. Herraiz came with the idea of working directly with the acquired list-mode data and without depending on the energy discrimination. This would avoid the need to perform any scanner modification and it could open the method to most current PET scanners. It was interesting to realize that almost all PET scanners record triple coincidences as a set of contiguous or almost contiguous double coincidences in the list-mode data. Multiple coincidences (several photons detected within the same time window) are not rejected, as it is considered that at least one of the double coincidences will be a correct one, and the other one will simply add some uniform background which can be corrected afterwards. Therefore, the list-mode data can be processed and identify and separate the triples from the double coincidences. This opened the possibility of performing dual-isotope imaging with clinical scanners. Some proof-of-concept studies were performed at Dr. Jose G. Venegas at Massachusetts General Hospital with the mCT PET scanner, although the accuracy of the method still had to be improved to be fully reliable.


This is when Joaquin L. Herraiz (who had just moved to Madrid to work at Complutense University of Madrid) met Jan Grimm in a “scientific speed-dating” organized at the European Molecular Imaging Conference. They started talking and realized that the multiplexed PET method could be very useful for the kind of research projects performed at Dr. Grimm’s Oncologic Molecular Imaging Lab at Memorial Sloan Kettering Cancer Center in New York, and it was an opportunity to continue improving the method to reach the required accuracy and reliability. They applied for a 5-year NIH grant to perform dual-isotope studies with the preclinical Inveon PET/CT scanner which was successfully funded after several submissions. This grant not only allowed performing interesting dual-isotope studies, but also gave the opportunity to finish the tools required to perform the mPET studies with speed and accuracy, especially with the development and use of the Monte Carlo simulator MCGPU-PET, in collaboration with Andreu Badal (FDA, USA).


At MSKCC Ph.D. Student Edwin C. Pratt and Dr. Grimm were excited to test out mPET on the Inveon PET/CT system. Since the system could record events in listmode format, Dr. Pratt was able to obtain separated dual-isotope images with the mPET method. The list-mode data from the studies acquired at MSKCC were sent to Madrid so that Ph.D. student Alejandro Lopez-Montes, Prof. Jose M. Udias and Ph.D. Joaquin L. Herraiz could perform the reconstruction and image separation. Alejandro Lopez-Montes who had started his Ph.D. on PET image reconstruction methods found this method quite exciting despite the additional complexity of the method with respect to standard PET acquisitions. For mPET the images could be quickly calibrated with a companion uniformity phantom scan in the same energy window for each isotope that would be used in future studies. The first experimental separation studies were done with the aid of a 3D printed mouse that Dr. Lukas Carter had made, and beautifully allowed the separation to be quantified and visually observed with a titration study of isotope mixtures at concentrations and volumes reasonable for preclinical studies. The group was excited to see the first separated images and the method was quantitative enough to move forward with other examples. The beauty of this work is that the PET acquisition was complete, and the reconstruction method for mPET could be redone anytime with better methods to improve separation.


The next example required two small molecule radiotracers, and since no 124I radiotracers were readily available and Grimm Lab at the time was interested in using trametinib, they first had to make a 124I trametinib isotopologue, which Dr. Edwin C. Pratt separately published in the Journal of Nuclear Medicine to really move this work further. The dual small molecule example with 18FDG and 124I-trametinib gave with mPET immediately gave a wealth of data, showing biological shifts in 18FDG on receptor tyrosine kinase therapy, and how well the tumor was blocked based on what part of the pathway was inhibited. The next example to test mPET involved two antibodies since 124I could be added to most antibodies with ease. In another alignment of events, Edwin Pratt was having dinner with another Ph.D student Michael J. Crowley and were discussing what targets could be explored for imaging immune responses. Looking at the state of T-cell exhaustion, Mike Crowley from the Vivek Mittal Lab, proposed trying to recapitulate a flow cytometry result in vivo. Ultimately, they did find differences in radiotracer uptake with tumor burden, though in hindsight much could be still optimized to fully elucidate T cell exhaustion. None the less, mPET imaging was possible.


Continuing to work with 124I-trametinib, Dr. Grimm’s Lab tried to load the inhibitor into ferumoxytol to show nanoparticle mediated delivery. Nanoparticle delivery has been a huge focus in the nanoparticle field, showing improved delivery to tissue using fluorescent imaging as the stand in for the drug of choice. Here Edwin C. Pratt really wanted to demonstrate with mPET that one could track both payload and delivery vehicle to the tumor. Everything in vitro qas well as prior experience (published in several papers) suggested a firm loading of trametinib into the particle yet mPET very quickly showed that the drug rapidly escaped the particle in this case. The lab has since tried several other methods of loading 124I-trametinib that ultimately show it diffuses very well especially at the picomolar amounts administered and is therefore not a good candidate for particle-based delivery methods. mPET gave an early read that nanoparticle loading was going to be difficult in vivo for this molecule, but an important lesion to learn. Previous unpublished work showed no clear effect of trametinib loaded ferumoxytol, but it was unclear if it was loading amount, stability, or just target resistance.


The group wanted the last main example to hit home that mPET could show delivery to tissue while confirming that the tissue was still positive. In talking with the Dr. Vladimir Ponomerov Lab also at MSKCC, they were interested in developing a PET reporter for their PSMA CAR-T. By designing a sodium iodide symporter into the CAR-T, they could track T Cell trafficking by bioluminescence as well as PET. Merging interests, Dr. Alessia Volpe, a Postdoctoral Fellow, and Dr. Edwin C. Pratt were able to combine traditional 68Ga-PSMA imaging with free 124I for CAR-T imaging simultaneously using mPET. The collaboration and results were thrilling as mPET could broadly separate the biodistribution of both tracers, and in particular, delivery of CAR-T was seen to be in the PSMA positive tumor as shown via BLI. In addition, it was exciting to see that the presence of CAR-Ts in the PSMA positive tumor was different from that of the 68Ga-PSMA alone, even though both were on the same intensity scale


The group was encouraged during the review process to better define the reconstruction performance of the method, and from these refinements the mPET methods was improved, and identified areas of weakness, and better validated the method overall. Perhaps the biggest finding from these comments was the positive bias observed in very low counts of the triple emitter, when the random triple rate exceeds the triple coincidences. This finding led Dr. Joaquin L. Herraiz to explore AI based learning methods to improve this reconstruction and could be a significant further improvement.


Overall, this work has tremendous potential in the future to improve imaging throughput or improve diagnostic detail, and what mPET can do is really built upon what new precise tracers can be developed. In particular isotopes such as 68Ga 124I, and 86Y have large positron flights, so using isotopes of lower energy like 89Zr and 52Mn will only help improve mPET imaging quality. In addition, there is hope this method can be used to screen not just for two known targets but enable investigational use to speed up imaging trials. In particular Dr. Joaquin L. Herraiz is excited about the use of AI in improving mPET and gaining higher order multiplexing capability. This work was incepted through meeting other investigators at a conference and over numerous dinners between many colleagues.


 It is the authors hope that most researchers can have the ability travel to imaging and oncology conferences alike and reach out to other researchers outside their direct field of interest. These collaborations helped identify new targets of interest and cross pollinate the best ideas.

Please sign in or register for FREE

If you are a registered user on Biotechnology and Bioengineering Community, please sign in