Precise measurement of scintillation events, which are light emissions from a scintillator material after ionizing radiation interaction, is crucial for various applications such as radiation source localization and radiation characterization. Information about the radiation source is encoded in the position and time of these events. Single Photon Avalanche Diodes (SPADs) offer high spatial and temporal resolution for single photon detection, making them attractive for this task. Existing methods using SiPM arrays provide excellent temporal resolution but suffer from low spatial resolution. Conversely, EMCCD cameras offer high spatial resolution but lack sufficient temporal resolution to discern individual particle interactions. This research aims to address this gap by exploring the capabilities of a high-resolution SPAD camera for imaging scintillation events in a thick, monolithic scintillator. The high spatial resolution allows for precise localization of the event's photons on the sensor minimizing noise accumulation from other pixels, while the high temporal resolution provides further denoising by separating events based on time of arrival. This novel approach, termed MOSSC (Monolithic Scintillator-SPAD Camera), holds promise for superior accuracy and versatility compared to existing methods, opening new possibilities for radiation detection and analysis.

Previous research on measuring scintillation events with SPAD arrays has been limited, often employing lensless designs with the SPAD array directly coupled to the scintillator's surface. These designs, while showcasing some success in PET applications and 2D particle tracking, suffer from limitations in spatial resolution and depth estimation capabilities. For instance, studies using low-resolution SPAD arrays in PET lack the spatial detail for accurate localization, while systems using scintillating fibers coupled to high-resolution SPAD arrays limit spatial resolution to the fiber diameter and preclude depth estimation. SiPM-based designs, while offering high temporal resolution, are severely constrained by their low spatial resolution due to a limited number of large-area readout channels. Although high-resolution cameras like EMCCD cameras exist, their low frame rates are insufficient to resolve individual interactions at high incidence rates. Other sensor technologies like flat-panel TFT arrays and Timepix chips offer high spatial resolution but lack the necessary temporal resolution or are highly specialized and less versatile. While LAPPDs offer competitive spatial and temporal resolutions, their current commercial availability is low and spatial resolution remains lower than advanced SPAD arrays. Notably, most previous work treated defocus blur as a problem to minimize, while this work leverages defocus blur as a tool for depth determination.

The MOSSC configuration uses a high-resolution SPAD camera coupled with a lens to focus light emitted from the scintillation events onto the sensor. The authors model each interaction as a point source of light, with its image forming a circle of confusion (CoC) whose diameter is related to the interaction's depth. An algorithm is developed for 3D interaction localization based on the image's defocus blur. The algorithm involves three key steps: 1. **Interaction detection:** Frames containing interactions are identified based on elevated counts within a defined region compared to a noise threshold determined from dark frames. 2. **Denoising and localization:** Dark counts are removed using a minimum spanning tree algorithm. A Gaussian mixture model (GMM) is fitted to the remaining photon coordinates to estimate the interaction's centroid and diameter. 3. **3D localization:** The interaction's 3D position is calculated using the circle of confusion equation (initially based on the thin lens approximation), relating the image diameter to depth, and perspective projection to determine lateral coordinates. To overcome limitations of the thin lens approximation imposed by the experimental setup's use of thick, high-curvature lenses, an optical calibration procedure is performed. This procedure creates a diameter-to-distance relationship, replacing the initial equation, allowing for accurate depth determination. Experimental hardware consists of a CsI(Tl) scintillator, a Co-60 gamma-ray source, a plano-convex lens, and the SwissSPAD2 camera. The gamma-ray source's position was systematically varied to evaluate the system's sensitivity to 3D shifts. Simulations were conducted using the Geant4 library to assess the MOSSC's 3D localization capability, considering factors like refraction blur, low-count bias, and different sensor types (SwissSPAD2 and Canon MS-500). Two localization methods were compared: GMM-Loc (Gaussian Mixture Model localization), used in experiments and simulations, and Exact-Loc, a simulation-only method using the exact photon coordinates. The simulations investigated various sources of bias (refraction, low-counts, denoising, and localization method) and assessed performance using spatial resolution (FWHM) and localization error.

Experiments demonstrated MOSSC's sensitivity to 3D shifts in the interaction's spatial distribution, with the distribution of measured interaction locations shifting consistently with the gamma-ray source's position both laterally and in depth. However, the magnitude of these shifts was relatively small, potentially due to the non-uniform sensitivity to interaction location along the depth dimension, the relatively large size of the gamma-ray source compared to the field of view, and the low signal and limited light collection. Simulations validated the MOSSC's 3D localization capability. Using the Exact-Loc method on noise-free data with the SwissSPAD2, the spatial resolution in x was around 0.2 mm, and in z was about 6-8 mm, demonstrating better lateral than depth resolution. Replicating experimental conditions (including noise and the denoising algorithm) significantly decreased the accuracy and spatial resolution in the z direction, emphasizing the limitations of the algorithm and the impact of noise in depth estimation. Simulations using the higher performance Canon MS-500 camera showed improved localization performance. An analysis of the uncertainty in the mean interaction position showed that the shifts observed in the experiments were significantly larger than the statistical uncertainty, confirming the system's 3D sensitivity. Finally, simulations explored the feasibility of using MOSSC as a monolithic Compton camera. While the SwissSPAD2 failed to accurately reconstruct the source position due to low light collection and resulting poor energy resolution, simulations with the Canon MS-500 achieved successful source reconstruction, demonstrating the potential of this approach with next-generation SPAD arrays.

The experimental results confirm the MOSSC system's sensitivity to 3D shifts in scintillation event distributions, but the observed shifts were smaller than expected. This can be attributed to several factors including non-uniform sensitivity along the depth, the relatively large source size compared to the field of view, and the low light levels resulting in the loss of some events. The simulations provided valuable insights into the system's performance and sources of error. They highlighted the significant impact of noise, limited photon counts, and the denoising and localization algorithms on the accuracy of depth estimation. The performance difference between the SwissSPAD2 and MS-500 simulations clearly illustrates the importance of sensor technology. The higher photon detection efficiency (PDE) of the MS-500 resulted in substantially better localization, demonstrating the potential for significantly improved performance with next-generation SPAD arrays. The successful Compton camera simulation with the MS-500 highlights a key application of the MOSSC system, leveraging the ability to resolve multiple interactions from individual gamma rays, a significant advantage over existing monolithic designs.

This research successfully demonstrated the feasibility of using a high-resolution SPAD camera for 3D imaging of scintillation events. The experimental results and simulations confirm the sensitivity of the MOSSC system to 3D shifts in the interaction locations and highlight the potential for accurate 3D localization, particularly with higher-performance SPAD arrays. The successful simulation of a monolithic Compton camera underscores the versatility and potential impact of MOSSC for advanced radiation detection applications. Future research should focus on optimizing the optical configuration, improving the denoising and localization algorithms, and exploring different scintillator materials and geometries to further enhance performance and reduce biases. The use of next-generation SPAD arrays with higher PDE and improved light collection will be crucial for realizing the full potential of MOSSC.

Several limitations of this study should be considered. The low light levels in the experiments necessitated operation at freezing temperatures, which is not practical for real-world applications. The relatively low photon detection efficiency of the SwissSPAD2 camera and the presence of dark counts introduced significant noise and decreased the accuracy of the localization, particularly in the depth dimension. The optical calibration procedure, while effective in correcting for the thin lens approximation, might introduce its own uncertainties. The chosen denoising and localization algorithms, while effective in reducing noise, may also introduce biases, especially in the depth estimation. Finally, the relatively small field of view of the experimental setup might limit the applicability of the proposed method to scenarios with closely positioned sources.