Point-of-care PET scanner offers interactive imaging

25 Mar 2019 Tami Freeman

Illustration showing the locations of the two PET detectors and the sampling pattern used to image the cylindrical phantom (the red circle). (Courtesy: Med. Phys. 10.1002/mp.13397)

Molecular imaging technologies such as PET are employed for in vivo diagnosis, evaluating disease progression and guiding therapeutic interventions. PET also offers the potential for imaging of novel theranostic ligands currently under development for personalized treatments. Standard whole-body PET systems, however, require installation in a large, dedicated scanning room and may not be the most cost-effective way to support the translation of novel PET ligands. Furthermore, it’s not always feasible to transfer a patient to the scanner.

What’s needed is a mobile and versatile PET scanner that can be brought to the patient for imaging at the bedside or in a treatment room. A team from Washington University in St Louisis developing just such a device: a point-of-care (POC) PET scanner for 3D tomographic molecular imaging. The compact POC-PET will use fast image reconstruction to give live feedback to the operator, enabling interactive scanning to optimize image quality (Med. Phys. 10.1002/mp.13397).Yuan‐Chuan Tai.

“The POC-PET may be used to provide near-real-time feedback in order to understand and optimize the delivery of the new theranostic ligands in their development phase,” says senior author Yuan‐Chuan Tai. “It may also be used to validate the delivery, or customize the dosing, of theranostic ligands to tailor the treatment protocol and support individualized medicine.”

Prototype device

Tai and colleagues constructed a proof-of-concept prototype POC-PET using two planar PET detectors, each comprising a 48×48 LYSO crystal array coupled to a photomultiplier tube. One detector is attached to a rotation stage and the second is mounted on a six-degrees-of-freedom robotic arm, enabling collection of coincidence events from multiple angles around the target.

The researchers employed a fast 3D image reconstruction engine — implemented on multiple graphics processing units (GPUs) — to read in the coincidence events and detector geometry and perform list-mode reconstruction using a simplified system matrix. The system matrix is computed on-the-fly, based on the changing detector position. As data are continuously collected and reconstructed, the system displays updated images in near real-time.

To characterize the prototype, the researchers first imaged a plane source between the detectors. Flood images showed that only the central 38×38 crystals in each array could be clearly resolved and used for subsequent imaging experiments. The PET detectors exhibited a coincidence resolving time (CRT) of about 740 ps FWHM for central crystals.

Next, they used the POC-PET to image a cylindrical phantom containing nine tumor spheres (with diameters of 3.6–11.4 mm) filled with 64Cu solution. They moved the two detectors to seven different locations and collected coincidence events from 27 sampling angles.

List-mode events reconstructed without TOF information contained visible artifacts, plus a hot spot outside of the phantom. In contrast, images reconstructed with TOF information had significantly reduced artifacts. The GPU-based image reconstruction took approximately 48 s for 10 iterations.(a) Inserts in the cylindrical phantom; (b) Sensitivity image of the prototype POC-PET system; (c) Image of the phantom reconstructed without TOF information, showing visible artefacts and a hot spot (red circle) outside of the phantom where the sensitivity is low; (d) Image reconstructed with TOF information. (Courtesy: Med. Phys. 10.1002/mp.13397)

Monte Carlo studies

Tai and colleagues also conducted Monte Carlo simulations of the POC-PET prototype, modeling the same set-up as in the experiment. Again, images reconstructed without TOF data exhibited more artifacts than those employing TOF information. Detectors with the fastest timing performance (coincidence resolving time (CRT) of 300 and 100 ps) produced the best image quality, with the smallest lesions (3.6 and 5 mm) only clearly visualized with a CRT of 100 ps.

To evaluate the proposed interactive scanning, the researchers divided the simulated list-mode data into six groups and reconstructed six images by adding one data set at a time, with or without TOF information (using a CRT of 300 ps). Four or five groups of data were sufficient to achieve a useful image, and reconstructions using TOF data showed fewer artifacts. The reconstruction time was approximately 65 s for ten iterations.

Tai notes that detectors with sub-300 ps CRT are now commercially available, and are becoming an industry standard for clinical PET. “It is not easy to achieve 100 ps CRT, yet,” he says. “However, our simulation suggests that 300 ps CRT will be sufficient to build a useful POC-PET system.”

Finally, the team simulated a body-sized torso phantom (containing 4–11 mm diameter tumors) imaged by a larger POC-PET system. The scanner included a front panel with 4×6 PET detector modules, each having 32×32 LYSO crystals, and a back panel with 3×8 PET detector modules, each made of 16×16 LYSO crystals.

The researchers imaged the phantom from four angles, reconstructing list-mode events from one sampling angle then adding an additional angle until all events were included. Image reconstruction took 55 s for a single angle and 108 s when using all of the data. They note that even with just three sampling angles, most tumors were resolved. Using four sampling angles clearly identified all tumors. These results suggest that a high-sensitivity POC-PET system could interactively image a body-size object in less than 7 min.

The researchers are now seeking funding to develop a product prototype. “We envision that a clinical POC-PET scanner may contain a panel that can be mounted behind a chair (or bed) and a maneuverable detector panel,” Tai tells Physics World. “It is significantly smaller than a standard body-sized PET scanner design and offers great potentials for future molecular imaging applications.”

Tami Freeman is the medical and biophysics editor for Physics World

FROM PHYSICSWORLD.COM 30/3/2019