Online program Overview Session: MIC-02

Abstract

Awake functional brain imaging is crucial for behavioral neuroscience, but conventional preclinical PET scanners typically require anesthesia, which suppresses brain glucose metabolism and limits natural behavioral studies. To address these limitations, we developed Gym-PET, a 75 g head-mounted wearable PET system optimized for awake running mice.

The wearable module features a compact detector ring (20.0 mm inner diameter) composed of 1 x 1 x 5 mm³ LYSO crystal arrays coupled to SiPMs, a short crystal design chosen to minimize parallax degradation under close-range conditions. To enable stable imaging during locomotion, the scanner is integrated with a customized treadmill platform using a lightweight suspension mechanism. Based on the NEMA NU-4 2008 standard, the system achieved 12.5% energy resolution, 0.8% peak sensitivity, and a high spatial resolution of 0.66 mm. Compared to conventional preclinical PET systems, Gym-PET demonstrated visibly improved delineation of small subcortical structures.

To validate the system's in vivo capabilities, dynamic 18F-FDG PET imaging was performed. The results successfully captured exercise-induced metabolic activation in the brainstem and thalamus of awake running mice. Furthermore, a three-day sleep deprivation study revealed the system's sensitivity to global metabolic suppression without anesthesia interference. Ultimately, the Gym-PET platform significantly reduces anesthesia-related metabolic interference, establishing a highly effective tool for high-resolution molecular neuroimaging in awake behavioral models.

Abstract

This study introduces Flexible On-demand Customized Ultra-resolution System for PET (FOCUS-PET), a novel technique designed to maximize spatial resolution and counting statistics within a predefined region of interest. The efficacy of FOCUS-PET was rigorously evaluated via Monte Carlo simulations and physical experiments using high-resolution add-on detector configurations. The simulated system employed an arc-shaped add-on detector comprised of a 4×2 module array, with each module housing 32×48 LYSO crystals (0.42×0.42×5.0 mm³). Sensitivity profiles were mapped using an axially stepped ²²Na point source, and a three-bed whole-body protocol was simulated utilizing a micro-Derenzo phantom. Simulation results demonstrated a 5.1% to 19.8% enhancement in absolute sensitivity. The system resolved 0.9-mm micro-rods, effectively surpassing native resolution limits, and improved the peak-to-valley ratio (PVR) and contrast recovery (CR) for 1.0-mm rods from 1.50 to 2.13 and 33.30% to 52.95%, respectively. Experimental validation was performed by integrating a flat-panel add-on detector—comprising two modules with 44×44 LYSO arrays (0.52×0.52×5.0 mm³)—into an ACTIVE 7 MAX preclinical scanner. Imaging of a physical Derenzo phantom was conducted to assess ultimate resolution and multi-bed stitching fidelity. Physical experiments corroborated the theoretical gains, with the flat-panel FOCUS-PET device consistently improving PVR (1.57 to 1.90) and CR (27.93% to 33.19%) for 1.25-mm hot rods compared to the baseline. In conclusion, FOCUS-PET significantly elevates system sensitivity and spatial resolution, providing a cost-effective pathway for ultra-high-resolution preclinical imaging and potential translation to clinical whole-body PET applications.

Abstract

High spatial resolution is essential to resolve small structures in the brain and to enable imaging of neurological pathways at the onset of diseases. A key challenge with ultra-high-resolution (UHR) PET scanners is the significant degradation of resolution that occurs beyond the central area of the field-of-view (FOV). The UHR Brain PET scanner was designed with short (12 mm) LYSO crystals to mitigate the parallax error off the FOV center, but at the expense of reduced sensitivity. To address these limitations, the LabPET II technology platform was upgraded to implement depth-of-interaction (DOI) measurement using longer (15 mm) dual LGSO phoswich detectors in the SAVANT. Simulation results indicate an estimated 40% gain in sensitivity, while maintaining the resolution below 2 mm FWHM at the edge of the brain. For DOI determination, the crystal identification between slow (43-48 ns decay time) and fast (30-35 ns decay time) LGSO scintillators uses a model-based dual-threshold time-over-threshold (ToT) discrimination technique, currently achieving better than 70% accuracy. The 6.5/8.5 mm length ratio of the top slow and bottom fast LGSO scintillators resulted from trade-offs between off-center spatial resolution and balanced coincidence detection efficiency of the two crystal layers. We report the initial experimental characterization of the partially assembled scanner (182 mm axial extent), which demonstrate the improvement in sensitivity and resolution uniformity across the radial FOV achieved using this simple DOI discrimination scheme. Point source and phantom measurements taken at various radial positions demonstrate the gain in spatial resolution compared to non-DOI measurements, though not reaching the resolution predicted by simulation. Further refinement of the crystal identification process using AI-based classification is expected to yield the forecasted benefits from simulation.  Preliminary results for imaging the human brain will be presented.

Abstract

Self-collimating SPECT (SC-SPECT) enables simultaneous high-resolution and high-sensitivity imaging using sparse-scintillator detector layers. Accurate system matrix (SM) estimation is essential but remains challenging in SC-SPECT, because dense point-source calibration is time-consuming and conventional modeling is limited by complex system physics, while geometry-dependent detector responses further complicate AI-based sparse-to-dense recovery. In this work, we propose a physics-guided sparse-to-dense SM modeling framework to recover the full SM from 10k point-source measurements. The framework uses a customized 3D encoder–decoder with anisotropic convolutions and modified down/up-sampling to preserve fine structures under limited sampling. Sparse response intensity, sampling positions, and detector angular information are encoded as four input channels, while LOR-directional profile and curvature-continuity losses enforce physically consistent recovery. The model is trained on simulated PPDFs, with ablation studies evaluating the input-channel, architecture, and loss designs. It is further validated using experimental SM-level metrics and MLEM reconstructions of hot-rod phantom and small-animal data. Results demonstrate improved PPDF recovery and reconstruction quality over interpolation-based SM estimation and a standard U-Net, while reducing the calibration time by over 97.5%.

Abstract

Dedicated brain PET systems are attracting renewed interest due to their potential to overcome key limitations of conventional whole-body PET, particularly limited sensitivity and spatial resolution for neurological imaging. In this work, we present BuckyPET, an innovative dedicated brain PET scanner inspired by fullerene-like geometry. The system is based on 234 regular hexagonal LYSO detector modules arranged to closely approximate a spherical configuration, arguably the most efficient geometry for brain PET imaging. Each detector is laser-engraved to create customizable optical barriers defining hexagonal scintillation pixels and is coupled to a hexagonal Hamamatsu MPPC with 61 individually read channels. 
Current clinical and preclinical NEMA standards rely on concepts such as sinograms, single-slice rebinning, axial axes, and transaxial planes, which are not directly applicable to spherical or highly enclosed geometries. We also discuss the need for a modified NEMA NU 2-2018-like protocol for non-cylindrical PET systems. We propose preserving the core performance metrics of the standard while reformulating the analysis in a geometry-independent framework.
Preliminary results demonstrate the high-sensitivity capabilities of BuckyPET, alongside a high noise-to-count ratio and spatial resolution in the range of 1.5-2 mm. Ongoing work focuses on fine-tuning the electronics and acquisition system and assessing robustness under different operating conditions. Thus, the final figures of merit are expected to further improve, while ongoing advances in the reconstruction framework are anticipated to push the spatial resolution beyond the 1 mm frontier. First studies for the clinical validation  are scheduled for summer 2026.