The effects of back-projection variants in BPF-like TOF PET reconstruction using CNN filtration - Based on simulated and clinical brain data.

BACKGROUND: The back-projection strategies such as confidence weighting (CW) and most likely annihilation position (MLAP) have been adopted into back-projection-and-filtering-like (BPF-like) deep reconstruction model and shown great potential on fast and accurate PET reconstruction. Although the two methods degenerate to an identical model at the time resolution of 0 ps, they represent two distinct approaches at the realistic time resolutions of current commercial systems. There is a lack of a systematic and fair assessment on these differences. PURPOSE: This work aims to analyze the impact of back-projection variants on CNN-based PET image reconstruction to find the most effective back-projection model, and ultimately contribute to accurate PET reconstruction. METHODS: Different back-projection strategies (CW and MLAP) and different angular view processing methods (view-summed and view-grouped) were considered, leading to the comparison of four back-projection variants integrated with the same CNN filtration model. Meanwhile, we investigated two strategies of physical effect compensation, either introducing pre-corrected data as the input or adding a channel of attenuation map to the CNN model. After training models separately on Monte-Carlo-simulated BrainWeb phantoms with full dose (events = 3×107), we tested them on both simulated phantoms and clinical brain scans with two dosage levels. For the performance assessment, peak signal-to-noise ratio (PSNR) and root mean square error (RMSE) were used to evaluate the pixel-wise error, structural similarity index (SSIM) to evaluate the structural similarity, and contrast recovery coefficient (CRC) in manually selected ROI to compare the region recovery. RESULTS: Compared to two MLAP-based histo-image reconstruction models, two CW-based back-projected image methods produced clearer, sharper, and more detailed images, from both simulated and clinical data. For angular view processing methods, view-grouped histo-image improved image quality, while view-grouped cwbp-image showed no advantage except for contrast recovery. Quantitative analysis on simulated data demonstrated that the view-summed cwbp-image model achieved the best PSNR, RMSE, SSIM, while the 8-view cwbp-image model achieved the best CRC in lesions and the white matter. Additionally, the multi-channel input model including the back-projection image and attenuation map was proved to be the most efficient and simplest method for compensating for physical effects for brain data. Applying Gaussian blur to the histo-image yielded images with limited improvement. All above results hold for both the half-dose and the full-dose cases. CONCLUSION: For brain imaging, the evaluation based on metrics PSNR, RMSE, SSIM, and CRC indicates that the view-summed CW-based back-projection variant is the most effective input for the BPF-like reconstruction model using CNN filtration, which can involve the attenuation map through an additional channel to effectively compensate for physical effects.

Design study of fully wearable high-performance brain PETs for neuroimaging in free movement.

A practical wearable brain PET scanner capable of dynamic neuroimaging during free bodily movement will enable potential breakthrough basic neuroscience studies and help develop imaging-based neurological diagnoses and treatments. Weight, brain coverage, and sensitivity are three fundamental technical obstacles in the development of Fully Wearable High-Performance (FWHP) brain PET scanners. The purpose of this study is to investigate the feasibility of building a FWHP brain PET using a limited volume of lutetium-yttrium oxyorthosilicate (LYSO) scintillator crystals. Six scanners, consisted of the same volume (2.66 kg) of LYSO scintillators with combinations of 2 different crystal pitches (3 mm and 1.5 mm) and 3 different crystal lengths (20 mm, 10 mm, and 5 mm), were simulated. The performances of the six scanners were assessed and compared with Siemen's HRRT brain PET and mCT whole-body PET, in terms of aperture, axial field of views (AFOV), sensitivity, spatial resolution, count rates, and image noise property. The time-of-flight (TOF) information was included in the image reconstruction to improve the effective sensitivity. The effects of the TOF was assessed by scanning a Jaszczak phantom and reconstructing images with the maximum likelihood expectation maximization (MLEM) algorithm with different timing settings (non-TOF, 500 ps, 200 ps, 100 ps and 50 ps Coincidence Time Resolution, CTR). The signal-noise ratio (SNR) of the images were assessed and compared with those of the HRRT scanner and mCT scanner. The results show that it is possible to construct a FWHP brain PET with better spatial resolution than the dedicated HRRT brain PET, comparable effective sensitivity (with 50 ∼ 100 ps CTR), and whole-brain coverage (23.7 cm inner diameter and 13.4 cm axial field of view) using 2.66 kg of LYSO.

A preclinical PET detector constructed with a monolithic scintillator ring.

This paper presents a unique preclinical positron emission tomography (PET) detector constructed with a monolithic scintillator ring (MSR) and two rings of silicon photomultipliers (SiPM). The inner diameter, outer diameter and length of the MSR were 48.5 mm, 58.5 mm, and 25.1 mm, respectively. The two SiPM rings, constructed with 46 SiPMs, were air-coupled to the two ends of the MSR detector. The center of gravity (COG) and artificial neural network (ANN) methods were adapted to decode the positions of the gamma interactions in the circumferential (θ) and axial (Z) directions, respectively. Collimating systems, consisting of a tungsten collimator and a high-precision displacement and rotating platform, were constructed to assess the decoding accuracies of the MSR detector in both θ and Z directions. The average intrinsic full-width half maximums (FWHMs) and mean absolute errors (MAEs) of the decoding accuracies were 0.94 mm and 0.33 mm in the circumferential direction, 2.45 mm and 1.08 mm in the axial direction. An energy resolution of 10.7% was measured at 511 keV. The scintillating photons generated by a pair of coincidence gamma photons overlap with each other, and cause circumferential parallax errors in the lines of response (LOR). The experimental results show that the average FWHM errors in the θ direction increased slightly from 0.94 mm to 1.14 mm when Δθ of the two single events was larger than 70°. The imaging performance of the MSR detector was also initially assessed with a Derenzo phantom filled with 18F-FDG. The rods with a diameter larger than 1.2 mm can be resolved. The energy resolutions were 12.3% at 511 keV (single events), and 11.4% at 1022 keV (coincidence events). We concluded that it is feasible to construct the high-performance preclinical PET scanners using one or multiple MSR detectors.

A depth encoding PET detector using four-crystals-to-one-SiPM coupling and light-sharing window method.

PURPOSE: Depth of interaction (DOI) decoding capability is of great importance for positron emission tomography (PET) requiring high resolution. In this study, we presented a novel low-cost DOI detector design with four crystals coupling to one SiPM, based on the method of rectangular light-sharing window (RLSW). A prototype detector was constructed, calibrated, and assessed using the methods of homogeneous radiation and flood map analysis. METHODS: The DOI detector was constructed with a 4 × 4 array of lutetium-yttrium oxyorthosilicate (LYSO) crystals (2.95 mm × 2.95 mm × 20 mm3 ), barium sulfate (BaSO4 ) reflectors, and optical glues. A RLSW 7 mm in height was deployed in the BaSO4 reflectors. A non-DOI detector with identical dimensions and without RLSW was also constructed for comparison. The light-output surface of the detector was air-coupled with a 4 × 4 array of SiPMs (3 mm × 3 mm2 ). The signals generated from the 16 SiPMs were read out by a custom-designed electronic system, and the signals from four adjacent 3 mm SiPMs were summed into one signal to emulate a 2 × 2 array of 6 mm SiPMs. The RLSW caused the DOI-related position shifts of the crystal spots in the flood map. A homogeneous radiation method was used to establish the transfer functions to convert the spot shifts measured from the flood map into DOI measurements. The accuracy of the DOI measurements was assessed with data acquired using the conventional collimated radiation method. RESULTS: All 16 crystals are distinctly separated from each other in the flood map. Twelve crystals, including four central crystals and eight edge crystals, have the DOI capability. The full width half maximum (FWHM) of the DOI measurements of the central crystals and the edge crystals are 3.06 ± 0.08 and 3.79 ± 0.15 mm, respectively, for the configuration with four crystals coupling to one SiPM. By contrast, the FWHMs (3.98 ± 0.16 and 5.12 ± 0.38 mm, respectively) are slightly worse for the configuration with one crystal coupling to one SiPM. The average and standard deviation (STD) of the FWHM energy resolutions of the DOI detector and non-DOI detector were 10.2% ± 0.7% and 10.7% ± 1.7%, respectively. Their FWHM coincidence timing resolutions were 197.0 ± 9.6 and 206.4 ± 13.3 ps, respectively. The RLSW had no significant impact on the energy resolutions and timing resolutions of the DOI detector. CONCLUSIONS: The novel four-crystals-to-one-SiPM coupling technology is a cost-efficient approach to construct high-performance detector modules with DOI capability. The methods of homogeneous radiation and flood map analysis are easy to perform and of good performance. Those methods can be adapted in the clinic PET scanners to enable the capability of DOI measurements.

Developing a 'multiPatchPET' system in GATE for a PET system design with irregular geometries.

This work modified the commonly used Monte Carlo tool package GATE by developing a new 'multiPatchPET' system so that GATE users can easily simulate PET systems with irregular geometries. The motivation was to design a brain PET scanner with high sensitivity. It is known that compact PET scanners with a large solid coverage angle can achieve high sensitivity with fewer scintillation detectors, and thus have the potential to provide better image quality in brain PET imaging than conventional ring PET scanners. However, considering a straightforward example with the largest possible solid angle, a spherical PET scanner is hard to manufacture. A more practical alternative would be a sphere-like polyhedral PET scanner with flat detector patches. Moreover, when monolithic scintillators are chosen to construct these flat detector modules, detection efficiency is further improved. Thus, we plan to design a sphere-like polyhedral PET scanner made up of monolithic scintillators. Unfortunately, in our design study, we found that simulating such a scanner with the latest GATE version (8.0) was not trivial, since no predefined systems could be used. In this work we introduced a 'multiPatchPET' system to GATE, which we and other GATE users will be able to use to develop PET scanners with any irregular geometry and any shape of patch. To validate our modification, a single block detector and an mCT scanner were simulated via both the original 'ecat' system and the new 'multiPatchPET' system. The results show no difference in terms of the detecting efficiency and reconstruction image. Then we used the 'multiPatchPET' system to simulate an 86 surface polyhedral brain PET scanner. Compared with two cylindrical brain scanners, the polyhedral brain scanner shows a higher sensitivity and has fewer noisy images. Thus, it was proved that our modification, which is accessible to the nuclear imaging research community, equipped GATE with a powerful and user-friendly tool to simulate complex scanners with irregular patches easily.

Enhancing the Image Quality via Transferred Deep Residual Learning of Coarse PET Sinograms.

Increasing the image quality of positron emission tomography (PET) is an essential topic in the PET community. For instance, thin-pixelated crystals have been used to provide high spatial resolution images but at the cost of sensitivity and manufacture expense. In this paper, we proposed an approach to enhance the PET image resolution and noise property for PET scanners with large pixelated crystals. To address the problem of coarse blurred sinograms with large parallax errors associated with large crystals, we developed a data-driven, single-image super-resolution (SISR) method for sinograms, based on the novel deep residual convolutional neural network (CNN). Unlike the CNN-based SISR on natural images, periodically padded sinogram data and dedicated network architecture were used to make it more efficient for PET imaging. Moreover, we included the transfer learning scheme in the approach to process cases with poor labeling and small training data set. The approach was validated via analytically simulated data (with and without noise), Monte Carlo simulated data, and pre-clinical data. Using the proposed method, we could achieve comparable image resolution and better noise property with large crystals of bin sizes of thin crystals with a bin size from to . Our approach uses external PET data as the prior knowledge for training and does not require additional information during inference. Meanwhile, the method can be added into the normal PET imaging framework seamlessly, thus potentially finds its application in designing low-cost high-performance PET systems.

Simulation study of a high-performance brain PET system with dodecahedral geometry.

PURPOSE: In brain imaging, the spherical PET system achieves the highest sensitivity when the solid angle is concerned. However, it is not practical. In this work, we designed an alternative sphere-like scanner, the dodecahedral scanner, which has a high sensitivity in imaging and a high feasibility to manufacture. We simulated this system and compared the performance with a few other dedicated brain PET systems. METHODS: Monte Carlo simulations were conducted to generate data of the dedicated brain PET system with the dodecahedral geometry (11 regular pentagon detectors). The data were then reconstructed using the in-house developed software with the fully three-dimensional maximum-likelihood expectation maximization (3D-MLEM) algorithm. RESULTS: Results show that the proposed system has a high-sensitivity distribution for the whole field of view (FOV). With a depth-of-interaction (DOI) resolution around 6.67 mm, the proposed system achieves the spatial resolution of 1.98 mm. Our simulation study also shows that the proposed system improves the image contrast and reduces noise compared with a few other dedicated brain PET systems. Finally, simulations with the Hoffman phantom show the potential application of the proposed system in clinical applications. CONCLUSIONS: In conclusion, the proposed dodecahedral PET system is potential for widespread applications in high-sensitivity, high-resolution PET imaging, to lower the injected dose.

An energy-optimized collimator design for a CZT-based SPECT camera.

In single photon emission computed tomography, it is a challenging task to maintain reasonable performance using only one specific collimator for radio-tracers over a broad spectrum of diagnostic photon energies, since photon scatter and penetration in a collimator differ with the photon energy. Frequent collimator exchanges are inevitable in daily clinical SPECT imaging, which hinders throughput while subjecting the camera to operational errors and damage. Our objective is to design a collimator, which independent of the photon energy performs reasonably well for commonly used radiotracers with low- to medium-energy levels of gamma emissions. Using the Geant4 simulation toolkit, we simulated and evaluated a parallel-hole collimator mounted to a CZT detector. With the pixel-geometry-matching collimation, the pitch of the collimator hole was fixed to match the pixel size of the CZT detector throughout this work. Four variables, hole shape, hole length, hole radius/width and the source-to-collimator distance were carefully studied. Scatter and penetration of the collimator, sensitivity and spatial resolution of the system were assessed for four radionuclides including 57Co, 99m Tc, 123I and 111In, with respect to the aforementioned four variables. An optimal collimator was then decided upon such that it maximized the total relative sensitivity (TRS) for the four considered radionuclides while other performance parameters, such as scatter, penetration and spatial resolution, were benchmarked to prevalent commercial scanners and collimators. Digital phantom studies were also performed to validate the system with the optimal square-hole collimator (23 mm hole length, 1.28 mm hole width, 0.32 mm septal thickness) in terms of contrast, contrast-to-noise ratio and recovery ratio. This study demonstrates promise of our proposed energy-optimized collimator to be used in a CZT-based gamma camera, with comparable or even better imaging performance versus commercial collimators such as low-energy high resolution (LEHR) and medium energy general purpose (MEGP) collimators.

Design Studies of a CZT-based Detector Combined with a Pixel-Geometry-Matching Collimator for SPECT Imaging.

Single Photon Emission Computed Tomography (SPECT) suffers limited efficiency due to the need for collimators. Collimator properties largely decide the data statistics and image quality. Various materials and configurations of collimators have been investigated in many years. The main thrust of our study is to evaluate the design of pixel-geometry-matching collimators to investigate their potential performances using Geant4 Monte Carlo simulations. Here, a pixel-geometry-matching collimator is defined as a collimator which is divided into the same number of pixels as the detector's and the center of each pixel in the collimator is a one-to-one correspondence to that in the detector. The detector is made of Cadmium Zinc Telluride (CZT), which is one of the most promising materials for applications to detect hard X-rays and γ-rays due to its ability to obtain good energy resolution and high light output at room temperature. For our current project, we have designed a large-area, CZT-based gamma camera (20.192 cm×20.192 cm) with a small pixel pitch (1.60 mm). The detector is pixelated and hence the intrinsic resolution can be as small as the size of the pixel. Materials of collimator, collimator hole geometry, detection efficiency, and spatial resolution of the CZT detector combined with the pixel-matching collimator were calculated and analyzed under different conditions. From the simulation studies, we found that such a camera using rectangular holes has promising imaging characteristics in terms of spatial resolution, detection efficiency, and energy resolution.