No abstract

Reducing radiation dose is important for PET imaging. However, reducing injection doses causes increased image noise and low signal-to-noise ratio (SNR), subsequently affecting diagnostic and quantitative accuracies. Deep learning methods have shown a great potential to reduce the noise and improve the SNR in low dose PET data.In this work, we comprehensively investigated the quantitative accuracy of small lung nodules, in addition to visual image quality, using deep learning based denoising methods for oncological PET imaging. We applied and optimized an advanced deep learning method based on the U-net architecture to predict the standard dose PET image from 10% low-dose PET data. We also investigated the effect of different network architectures, image dimensions, labels and inputs for deep learning methods with respect to both noise reduction performance and quantitative accuracy. Normalized mean square error (NMSE), SNR, and standard uptake value (SUV) bias of different nodule regions of interest (ROIs) were used for evaluation.Our results showed that U-net and GAN are superior to CAE with smaller SUV mean and SUV max bias at the expense of inferior SNR. A fully 3D U-net has optimal quantitative performance compared to 2D and 2.5D U-net with less than 15% SUV mean bias for all the ten patients. U-net outperforms Residual U-net (r-U-net) in general with smaller NMSE, higher SNR and lower SUV max bias. Fully 3D U-net is superior to several existing denoising methods, including Gaussian filter, anatomical-guided non-local mean (NLM) filter, and MAP reconstruction with Quadratic prior and relative difference prior, in terms of superior image quality and trade-off between noise and bias. Furthermore, incorporating aligned CT images has the potential to further improve the quantitative accuracy in multi-channel U-net.We found the optimal architectures and parameters of deep learning based methods are different for absolute quantitative accuracy and visual image quality. Our quantitative results demonstrated that fully 3D U-net can both effectively reduce image noise and control bias even for sub-centimeter small lung nodules when generating standard dose PET using 10% low count down-sampled data.

Existing respiratory motion-correction methods are applied only to static PET imaging. We have previously developed an eventby-event respiratory motion-correction method with correlations between internal organ motion and external respiratory signals (INTEX). This method is uniquely appropriate for dynamic imaging because it corrects motion for each time point. In this study, we applied INTEX to human dynamic PET studies with various tracers and investigated the impact on kinetic parameter estimation. Methods: The use of 3 tracers-a myocardial perfusion tracer, 82 Rb (n = 7); a pancreatic β-cell tracer, 18 F-FP(+)DTBZ (n = 4); and a tumor hypoxia tracer, 18 F-fluoromisonidazole ( 18 F-FMISO) (n = 1)-was investigated in a study of 12 human subjects. Both rest and stress studies were performed for 82 Rb. The Anzai belt system was used to record respiratory motion. Three-dimensional internal organ motion in high temporal resolution was calculated by INTEX to guide event-by-event respiratory motion correction of target organs in each dynamic frame. Time-activity curves of regions of interest drawn based on endexpiration PET images were obtained. For 82 Rb studies, K 1 was obtained with a 1-tissue model using a left-ventricle input function. Rest-stress myocardial blood flow (MBF) and coronary flow reserve (CFR) were determined. For 18 F-FP(1)DTBZ studies, the total volume of distribution was estimated with arterial input functions using the multilinear analysis 1 method. For the 18 F-FMISO study, the net uptake rate K i was obtained with a 2-tissue irreversible model using a left-ventricle input function. All parameters were compared with the values derived without motion correction. Results: With INTEX, K 1 and MBF increased by 10% ± 12% and 15% ± 19%, respectively, for 82 Rb stress studies. CFR increased by 19% ± 21%. For studies with motion amplitudes greater than 8 mm (n 5 3), K 1 , MBF, and CFR increased by 20% ± 12%, 30% ± 20%, and 34% ± 23%, respectively. For 82 Rb rest studies, INTEX had minimal effect on parameter estimation. The total volume of distribution of 18 F-FP(1)DTBZ and K i of 18 F-FMISO increased by 17% ± 6% and 20%, respectively. Conclusion: Respiratory motion can have a substantial impact on dynamic PET in the thorax and abdomen. The INTEX method using continuous external motion data substantially changed parameters in kinetic modeling. More accurate estimation is expected with INTEX.

The authors demonstrate that the TEW method leads to overestimation in scatter and crosstalk for the CZT-based imaging system while the proposed scatter and crosstalk correction method can provide more accurate self-scatter and down-scatter estimations for quantitative single-radionuclide and dual-radionuclide imaging.

The detection of depth-of-interaction (DOI) is a critical detector capability to improve the PET spatial resolution uniformity across the field-of-view and will significantly enhance, in particular, small bore system performance for brain, breast, and small animal imaging. One promising technique of DOI detection is to use dual-ended-scintillator readout that uses two photon sensors to detect scintillation light from both ends of a scintillator array and estimate DOI based on the ratio of signals (similar to Anger logic). This approach needs a careful DOI function calibration to establish accurate relationship between DOI and signal ratios, and to recalibrate if the detection condition is shifted due to the drift of sensor gain, bias variations, or degraded optical coupling, etc. However, the current calibration method that uses coincident events to locate interaction positions inside a single scintillator crystal has severe drawbacks, such as complicated setup, long and repetitive measurements, and being prone to errors from various possible misalignments among the source and detector components. This method is also not practically suitable to calibrate multiple DOI functions of a crystal array. To solve these problems, a new method has been developed that requires only a uniform flood source to irradiate a crystal array without the need to locate the interaction positions, and calculates DOI functions based solely on the uniform probability distribution of interactions over DOI positions without knowledge or assumption of detector responses. Simulation and experiment have been studied to validate the new method, and the results show that the new method, with a simple setup and one single measurement, can provide consistent and accurate DOI functions for the entire array of multiple scintillator crystals. This will enable an accurate, simple, and practical DOI function calibration for the PET detectors based on the design of dual-ended-scintillator readout. In addition, the new method can be generally applied to calibrating other types of detectors that use the similar dual-ended readout to acquire the radiation interaction position.