CBCT-DRRs superior to CT-DRRs for target-tracking applications for pancreatic SBRT.

Objective.In current radiograph-based intra-fraction markerless target-tracking, digitally reconstructed radiographs (DRRs) from planning CTs (CT-DRRs) are often used to train deep learning models that extract information from the intra-fraction radiographs acquired during treatment. Traditional DRR algorithms were designed for patient alignment (i.e.bone matching) and may not replicate the radiographic image quality of intra-fraction radiographs at treatment. Hypothetically, generating DRRs from pre-treatment Cone-Beam CTs (CBCT-DRRs) with DRR algorithms incorporating physical modelling of on-board-imagers (OBIs) could improve the similarity between intra-fraction radiographs and DRRs by eliminating inter-fraction variation and reducing image-quality mismatches between radiographs and DRRs. In this study, we test the two hypotheses that intra-fraction radiographs are more similar to CBCT-DRRs than CT-DRRs, and that intra-fraction radiographs are more similar to DRRs from algorithms incorporating physical models of OBI components than DRRs from algorithms omitting these models.Approach.DRRs were generated from CBCT and CT image sets collected from 20 patients undergoing pancreas stereotactic body radiotherapy. CBCT-DRRs and CT-DRRs were generated replicating the treatment position of patients and the OBI geometry during intra-fraction radiograph acquisition. To investigate whether the modelling of physical OBI components influenced radiograph-DRR similarity, four DRR algorithms were applied for the generation of CBCT-DRRs and CT-DRRs, incorporating and omitting different combinations of OBI component models. The four DRR algorithms were: a traditional DRR algorithm, a DRR algorithm with source-spectrum modelling, a DRR algorithm with source-spectrum and detector modelling, and a DRR algorithm with source-spectrum, detector and patient material modelling. Similarity between radiographs and matched DRRs was quantified using Pearson's correlation and Czekanowski's index, calculated on a per-image basis. Distributions of correlations and indexes were compared to test each of the hypotheses. Distribution differences were determined to be statistically significant when Wilcoxon's signed rank test and the Kolmogorov-Smirnov two sample test returnedp≤ 0.05 for both tests.Main results.Intra-fraction radiographs were more similar to CBCT-DRRs than CT-DRRs for both metrics across all algorithms, with allp≤ 0.007. Source-spectrum modelling improved radiograph-DRR similarity for both metrics, with allp< 10-6. OBI detector modelling and patient material modelling did not influence radiograph-DRR similarity for either metric.Significance.Generating DRRs from pre-treatment CBCT-DRRs is feasible, and incorporating CBCT-DRRs into markerless target-tracking methods may promote improved target-tracking accuracies. Incorporating source-spectrum modelling into a treatment planning system's DRR algorithms may reinforce the safe treatment of cancer patients by aiding in patient alignment.

Evaluation of deep learning based implanted fiducial markers tracking in pancreatic cancer patients.

Real-time target position verification during pancreas stereotactic body radiation therapy (SBRT) is important for the detection of unplanned tumour motions. Fast and accurate fiducial marker segmentation is a Requirement of real-time marker-based verification. Deep learning (DL) segmentation techniques are ideal because they don't require additional learning imaging or prior marker information (e.g., shape, orientation). In this study, we evaluated three DL frameworks for marker tracking applied to pancreatic cancer patient data. The DL frameworks evaluated were (1) a convolutional neural network (CNN) classifier with sliding window, (2) a pretrained you-only-look-once (YOLO) version-4 architecture, and (3) a hybrid CNN-YOLO. Intrafraction kV images collected during pancreas SBRT treatments were used as training data (44 fractions, 2017 frames). All patients had 1-4 implanted fiducial markers. Each model was evaluated on unseen kV images (42 fractions, 2517 frames). The ground truth was calculated from manual segmentation and triangulation of markers in orthogonal paired kV/MV images. The sensitivity, specificity, and area under the precision-recall curve (AUC) were calculated. In addition, the mean-absolute-error (MAE), root-mean-square-error (RMSE) and standard-error-of-mean (SEM) were calculated for the centroid of the markers predicted by the models, relative to the ground truth. The sensitivity and specificity of the CNN model were 99.41% and 99.69%, respectively. The AUC was 0.9998. The average precision of the YOLO model for different values of recall was 96.49%. The MAE of the three models in the left-right, superior-inferior, and anterior-posterior directions were under 0.88 ± 0.11 mm, and the RMSE were under 1.09 ± 0.12 mm. The detection times per frame on a GPU were 48.3, 22.9, and 17.1 milliseconds for the CNN, YOLO, and CNN-YOLO, respectively. The results demonstrate submillimeter accuracy of marker position predicted by DL models compared to the ground truth. The marker detection time was fast enough to meet the requirements for real-time application.

A validated Geant4 model of a whole-body PET scanner with four-layer DOI detectors.

The purpose of this work is to develop a validated Geant4 simulation model of a whole-body prototype PET scanner constructed from the four-layer depth-of-interaction detectors developed at the National Institute of Radiological Sciences, National Institutes for Quantum and Radiological Science and Technology, Japan. The simulation model emulates the behaviour of the unique depth of interaction sensing capability of the scanner without needing to directly simulate optical photon transport in the scintillator and photodetector modules. The model was validated by evaluating and comparing performance metrics from the NEMA NU 2-2012 protocol on both the simulated and physical scanner, including spatial resolution, sensitivity, scatter fraction, noise equivalent count rates and image quality. The results show that the average sensitivities of the scanner in the field-of-view were 5.9 cps kBq-1 and 6.0 cps kBq-1 for experiment and simulation, respectively. The average spatial resolutions measured for point sources placed at several radial offsets were 5.2± 0.7 mm and 5.0± 0.8 mm FWHM for experiment and simulation, respectively. The peak NECR was 22.9 kcps at 7.4 kBq ml-1 for the experiment, while the NECR obtained via simulation was 23.3 kcps at the same activity. The scatter fractions were 44% and 41.3% for the experiment and simulation, respectively. Contrast recovery estimates performed in different regions of a simulated image quality phantom matched the experimental results with an average error of -8.7% and +3.4% for hot and cold lesions, respectively. The results demonstrate that the developed Geant4 model reliably reproduces the key NEMA NU 2-2012 performance metrics evaluated on the prototype PET scanner. A simplified version of the model is included as an advanced example in Geant4 version 10.5.

Design study of a brain-dedicated time-of-flight PET system with a hemispherical detector arrangement.

Time-of-flight (TOF) is now a standard technology for positron emission tomography (PET), but its effective use for small diameter PET systems has not been studied well. In this paper, we simulated a brain-dedicated TOF-PET system with a hemispherical detector arrangement. We modeled a Hamamatsu TOF-PET module (C13500-4075LC-12) with 280 ps coincidence resolving time (CRT), in which a 12 × 12 array of multi pixel photon counters (MPPCs) is connected to a lutetium fine silicate (LFS) crystal array of 4.1 × 4.1 mm2 cross section each, based on one-to-one coupling. On the other hand, spatial resolution degradation due to the parallax error should be carefully addressed for the small diameter PET systems. The ideal PET detector would have both depth-of-interaction (DOI) and TOF capabilities, but typical DOI detectors that are based on light sharing tend to degrade TOF performance. Therefore, in this work, we investigated non-DOI detectors with an appropriate crystal length, which was a compromise between suppressed parallax error and decreased sensitivity. Using GEANT4, we compared two TOF detectors, a 20 mm long non-DOI and a 10 mm long non-DOI, with a non-TOF, 4-layer DOI detector with a total length of 20 mm (i.e. 5 × 4 mm). We simulated a contrast phantom and evaluated the relationship between the contrast recovery coefficient (CRC) and the noise level (the coefficient of variation, COV) for reconstructed images. The 10 mm long non-DOI, which reduces the parallax error at the cost of sensitivity loss, showed better imaging quality than the 20 mm long non-DOI. For example, the CRC value of a 10 mm hot sphere at COV = 20% was 72% for the 10 mm long non-DOI, which was 1.2 times higher than that of the 20 mm long non-DOI. The converged CRC values for the 10 mm long non-DOI were almost equivalent to those of the non-TOF 4-layer DOI, and the 10 mm long non-DOI converged faster than the non-TOF 4-layer DOI did. Based on the simulation results, we evaluated a one-pair prototype system of the TOF-PET detectors with 10 mm crystal length, which yielded the CRT of 250 ± 8 ps. In summary, we demonstrated support for feasibility of the brain-dedicated TOF-PET system with the hemispherical detector arrangement.

Comparative study of alternative Geant4 hadronic ion inelastic physics models for prediction of positron-emitting radionuclide production in carbon and oxygen ion therapy.

The distribution of fragmentation products predicted by Monte Carlo simulations of heavy ion therapy depend on the hadronic physics model chosen in the simulation. This work aims to evaluate three alternative hadronic inelastic fragmentation physics options available in the Geant4 Monte Carlo radiation physics simulation framework to determine which model most accurately predicts the production of positron-emitting fragmentation products observable using in-beam PET imaging. Fragment distributions obtained with the BIC, QMD, and INCL + + physics models in Geant4 version 10.2.p03 are compared to experimental data obtained at the HIMAC heavy-ion treatment facility at NIRS in Chiba, Japan. For both simulations and experiments, monoenergetic beams are applied to three different block phantoms composed of gelatin, poly(methyl methacrylate) and polyethylene. The yields of the positron-emitting nuclei 11C, 10C and 15O obtained from simulations conducted with each model are compared to the experimental yields estimated by fitting a multi-exponential radioactive decay model to dynamic PET images using the normalised mean square error metric in the entrance, build up/Bragg peak and tail regions. Significant differences in positron-emitting fragment yield are observed among the three physics models with the best overall fit to experimental 12C and 16O beam measurements obtained with the BIC physics model.

Investigation of spatial resolution improvement by use of a mouth-insert detector in the helmet PET scanner.

The dominant factor limiting the intrinsic spatial resolution of a positron emission tomography (PET) system is the size of the crystal elements in the detector. To increase sensitivity and achieve high spatial resolution, it is essential to use advanced depth-of-interaction (DOI) detectors and arrange them close to the subject. The DOI detectors help maintain high spatial resolution by mitigating the parallax error caused by the thickness of the scintillator near the peripheral regions of the field-of-view. As an optimal geometry for a brain PET scanner, with high sensitivity and spatial resolution, we proposed and developed the helmet-chin PET scanner using 54 four-layered DOI detectors consisting of a 16 × 16 × 4 array of GSOZ scintillator crystals with dimensions of 2.8 × 2.8 × 7.5 mm3. All the detectors used in the helmet-chin PET scanner had the same spatial resolution. In this study, we conducted a feasibility study of a new add-on detector arrangement for the helmet PET scanner by replacing the chin detector with a segmented crystal cube, having high spatial resolution in all directions, which can be placed inside the mouth. The crystal cube (which we have named the mouth-insert detector) has an array of 20 × 20 × 20 LYSO crystal segments with dimensions of 1 × 1 × 1 mm3. Thus, the scanner is formed by the combination of the helmet and mouth-insert detectors, and is referred to as the helmet-mouth-insert PET scanner. The results show that the helmet-mouth-insert PET scanner has comparable sensitivity and improved spatial resolution near the center of the hemisphere, compared to the helmet-chin PET scanner.

Simulation study comparing the helmet-chin PET with a cylindrical PET of the same number of detectors.

There is a growing interest in developing brain PET scanners with high sensitivity and high spatial resolution for early diagnosis of neurodegenerative diseases and studies of brain functions. Sensitivity of the PET scanner can be improved by increasing the solid angle. However, conventional PET scanners are designed based on a cylindrical geometry, which may not be the most efficient design for brain imaging in terms of the balance between sensitivity and cost. We proposed a dedicated brain PET scanner based on a hemispheric shape detector and a chin detector (referred to as the helmet-chin PET), which is designed to maximize the solid angle by increasing the number of lines-of-response in the hemisphere. The parallax error, which PET scanners with a large solid angle tend to have, can be suppressed by the use of depth-of-interaction detectors. In this study, we carry out a realistic evaluation of the helmet-chin PET using Monte Carlo simulation based on the 4-layer GSO detector which consists of a 16 × 16 × 4 array of crystals with dimensions of 2.8 × 2.8 × 7.5 mm3. The purpose of this simulation is to show the gain in imaging performance of the helmet-chin PET compared with the cylindrical PET using the same number of detectors in each configuration. The sensitivity of the helmet-chin PET evaluated with a cylindrical phantom has a significant increase, especially at the top of the (field-of-view) FOV. The peak-NECR of the helmet-chin PET is 1.4 times higher compared to the cylindrical PET. The helmet-chin PET provides relatively low noise images throughout the FOV compared to the cylindrical PET which exhibits enhanced noise at the peripheral regions. The results show the helmet-chin PET can significantly improve the sensitivity and reduce the noise in the reconstructed images.

Simulation study of a D-shape PET scanner for improved sensitivity and reduced cost in whole-body imaging.

Much research effort is being made to increase the sensitivity and improve the imaging performance of positron emission tomography (PET) scanners. Conventionally, sensitivity can be increased by increasing the number of detector rings in the axial direction (but at high cost) or reducing the diameter of the scanner (with the disadvantages of reducing the space for patients and degrading the spatial resolution due to the parallax error). In this study, we proposed a PET scanner with a truncated ring and an array of detectors that can be arranged in a straight line below the bed. We called this system 'D-PET' as it resembles the letter 'D' when it is rotated by 90° in the counterclockwise direction. The basic design idea was to cut the unused space under the patient's bed; this area is usually not in use in clinical diagnosis. We conducted Monte Carlo simulations of the D-PET scanner and compared its performance with a cylindrical PET scanner. The scanners were constructed from 4-layer depth-of-interaction detectors which consisted of a 16 × 16 × 4 LYSO crystal array with dimensions of 2.85 × 2.85 × 5 mm3. The results showed that the D-PET had an increase in sensitivity and peak-NECR of 30% and 18%, respectively. The D-PET had low noise in the reconstructed images throughout the field-of-view compared to the cylindrical PET. These were achieved while keeping sufficient space for the patient, and also without a severe effect on the spatial resolution. Furthermore, the number of detectors (and hence the cost) of the D-PET scanner was reduced by 12% compared to the cylindrical PET scanner.

Pre-computed system matrix calculation based on a piece-wise method for PET.

A system matrix (SM) is the basic component of iterative image reconstruction algorithms. Calculation of the SM needs a considerable amount of time due to an enormous number of lines of response (LORs) being modeled. In this study, we developed a technique based on a piece-wise calculation method in which symmetry and further division of the voxels are applied. The detector response function for all detectable pairs of photons along certain LORs originating from each voxel is calculated analytically. The total number of LORs in 300 × 300 × 120 voxels (with 2 × 2 × 2 mm(3)) is ~44 billion, and the SM was calculated by the use of three different computers independently; the calculation time was 5 h. The SM took 5 days when calculated by the use of the conventional method (where symmetry and the piece-wise method are not used). The sensitivity correction factor was stored; it had a size of 42 MB in a four-byte computer memory.