Online program Overview Session: MIC-06

Abstract

Single-photon emission computed tomography (SPECT) pinhole collimators trade isotropic resolution against detector coverage: circular apertures cover only 78.5% of a square detector. Rectangular pinholes (rectholes) fill the detector but introduce anisotropic resolution. Lofthole collimators linearly interpolate between a circular aperture and a rectangular exit; their non-developable lateral surfaces complicate fabrication and geometric calibration. We present the minkhole collimator, a knife-edge geometry whose half-frustum equals the Minkowski sum of a circular waist and a rectangular exit. The half-frustum boundary decomposes exactly into ten analytic patches: two caps, four ruled side planes, and four corner quadrics. Convexity of each half-frustum bounds a ray to at most two boundary intersections, giving a closed-form ray tracer of six linear and four quadratic solves with no iteration. We implement the 10-patch intersection and compare the minkhole to pinhole, recthole, and lofthole collimators on a common geometry. At matched sensitivity and field of view, minkhole projections are more symmetric than the recthole's and show less diagonal elongation than the lofthole's, while fully covering the detector. Its piecewise-developable surface (planar and sheared-conical patches) admits a compact analytic form that simplifies calibration, making the minkhole a practical building block for multi-aperture SPECT.

Abstract

The aim of this study was to develop and evaluate a GATE Monte Carlo model of the ring CZT based VERITON SPECT-CT system against experimental data. System spatial resolution was evaluated using planar images with ^99mTc point-like sources. Good agreement was obtained between both experimental and simulated data with consistent values within one sigma, except for the studied distances closest to the collimator surface (5 and 7 cm) where relative discrepancies of 25% and 7% were observed along X-axis. A tomographic acquisition, corresponding to 4 gantry rotations and 60 swivel positions per detector head, using the NEMA IEC body phantom was simulated within GATE 10. A qualitative comparison of profiles showed good agreement between both reconstructions.
We plan to further evaluate this model regarding quantitative imaging and extend it to additional radionuclides. Such a model could have a significant impact on the optimisation of acquisition protocols, image reconstruction and system design.

Abstract

For metastatic lymph-node localization in esophageal cancer surgery with ¹⁸F-FDG injection, we have been developing an intraoperative positron emission counter (PEC). This device employs two BGO scintillators (∼10×5×15 mm³) on the tip of a forceps and it provides coincidence detection. The current design of the PEC is not position sensitive; however, quantitative radioactivity measurement requires recovery of the source distribution within the compact field of view (FOV) of the PEC, because position-dependent detection efficiency renders the integral count rate ambiguous. In this paper, we investigate the feasibility of source distribution visualization on a position-sensitive PEC variant featuring 2×2 segmentation of each scintillator (yielding 16 lines of response, LORs) together with a machine-learning (ML) decoder mapping these 16 LORs to a source distribution. Specifically, we designed a multi-layer perceptron (MLP) with a spatial-softmax head defined over the FOV. The MLP was trained with Kullback-Leibler (KL)-divergence supervision on random-walk phantoms (each yielding a 16-LOR expected count vector via Geant4 simulation, Poisson-sampled at a 100-count budget), with mixed single- and synthetic 2-source training samples. The pitch of the FOV reconstruction grid was 0.5 mm. On held-out single-source test phantoms, the predicted distribution provides the source center-of-gravity (CoG) with a median 3D Euclidean mean-absolute-error (MAE) of 0.73 mm and the per-axis source extent within ≤16% (sigma MAE 0.13–0.24 mm vs. ground-truth 1.2–1.5 mm). On disjoint 2-source pairs, 90.1% of the detected peaks lie within 2.5 mm of a true source (per-detected-peak median 1.40 mm; per-true-source median 1.58 mm). These results indicate that a position-sensitive PEC with an MLP-based decoder achieves not only the source localization but also the source distribution visualization from only 16 LOR counts.

Abstract

Metascintillator systems composed of LYSO:Ce and 2D organic lead halide perovskites were recently measured in the laboratory with high end electronics. The timing behavior of this metascintillator rendered two dominant subsets: 44% of fast, perovskite-led and shared events at 46 ps and 56% of slow, LYSO:Ce events at 110 ps. As this design reaches ultrafast ToF regimes, an opportunity to better understand the effect of such CTR on final image quality arises. The possibility of adapting this configuration to a photonic-driven semi-monolithic metascintillator also offers very good depth-of-interaction discrimination, at 4 mm. To benchmark this detector architecture, a brain PET scanner was designed in GATE, with a Hoffman brain and Derenzo phantoms for benchmarking. A series of reconstruction approaches, for instance using the fast subset as priors or weighing in on the different subsets been performed, but the drive for this exercise is to invite the reconstruction community to experiment with and demonstrate the potential of new generation high end ToF detectors for medical imaging.

Abstract

Last year, we developed a sub-0.5 mm resolution positron emission tomography (PET) system using staggered 3-layer depth-of-interaction (DOI) detectors for mouse brain imaging. However, the crystal pitch of 0.8 mm limited the spatial resolution to 0.45 mm, which sparked our pursuit of the ultimate resolution. In this study, we investigate the spatial resolution of a newly proposed mouse brain PET system that uses a reduced crystal pitch as we move towards sub-0.4 mm resolution. The proposed system provides 48 mm inner diameter and 22.6 mm axial coverage. It consists of 32 DOI detectors with 2-ring geometry. Each DOI detector has staggered 3-layer LYSO crystals (crystal pitch=0.6 mm) with a crystal layer thickness of 3+3+5 mm, thereby providing a 11 mm total thickness. We used GATE simulation to investigate the performance of the proposed PET design. For spatial resolution evaluation, an NEMA ²²Na source (D=0.25 mm) was scanned from the field-of-view center to the 15 mm radial offset position and it was reconstructed by using an FBP algorithm. For imaging performance evaluation, a resolution phantom having six different rod diameters was used. The PET images were reconstructed by using the OSEM algorithm. Compared to results of our previous work, the spatial resolution was improved from 0.75±0.02 mm to 0.55±0.07 mm. Unlike the previous work, the 0.4 mm rods of the resolution phantom were resolved with a valley-to-peak ratio of 0.763. In conclusion, the simulation showed that our proposed mouse brain PET system provided 0.4 mm resolution in resolution phantom imaging. As a next step, we plan to develop a prototype mouse brain PET scanner to achieve 0.4 mm resolution experimentally.