Abstract:572 Objectives: Attenuation correction is necessary in order to perform accurate qualitative or quantitative analysis for single photon emission computed tomography (SPECT). Hybrid SPECT/CT systems equipped with CT scanners can provide direct measurement of photon attenuation, but are substantially more expensive than conventional SPECT systems and often require larger imaging rooms, additional shielding, and relatively complicated acquisition protocols. Many current SPECT-only systems do not support transmission scanning and therefore are susceptible to attenuation artifacts. Where available, the use of transmission scanning also increases radiation doses to the patient and significant artifacts could occur due to mismatches between the emission and transmission scans as a result of patient motion. Due to all these reasons, we propose a method that estimates an attenuation map (ATTMAP) directly from SPECT emission data using deep neural networks. Methods: It has been shown in previous studies that both photopeak photons and scatter photons contain information that could help estimate the attenuation distribution. In our proposed method, images reconstructed from photopeak window (126 keV-155 keV) and scatter window (114 keV-126 keV) are concatenated as a multi-channel image and fed into the deep neural networks to generate synthetic attenuation coefficient images. A 3D U-Net deep convolutional neural network (CNN) with a Generative Adversarial Network (GAN) training strategy was used to estimate attenuation maps from emission images. The GAN used an additional 3D CNN network as a discriminator to enforce the generator’s output to be consistent with the ground truth attenuation maps as much as possible. An image gradient difference term was also added to the loss function to retain the sharpness of the generated attenuation maps. The training set includes 40 human subjects with both cardiac SPECT with 99mTc-tetrofosmin and attenuation CT scans, and the testing set includes 8 subjects not involved in the network training. The SPECT and CT images were acquired from a GE NM/CT 850 SPECT/CT scanner. The synthetic attenuation maps were compared with the true attenuation maps regarding both global normalized Mean Absolute Error (NMAE=MAE(synthetic)/[max(true)-min(true)]) and localized region of interest (ROI) absolute percentage error (|(roi_mean(synthetic)-roi_mean(true))/roi_mean(true)|) in left ventricle (LV) myocardium (121.8±30.0 cm3) and LV blood pool (40.7±7.5 cm3) ROIs. The localized absolute percentage error was also calculated for attenuation corrected SPECT reconstruction images with both true and synthetic attenuation maps. Results: The proposed method successfully generated accurate synthetic attenuation maps close to the true attenuation map, both qualitatively and quantitatively. The SPECT reconstructed images corrected using the true attenuation map and synthetic attenuation map are almost identical, whereas obvious attenuation artifacts can be observed in the non-attenuation corrected images. The global NMAE of the synthetic attenuation maps across the 8 testing subjects were 3.4%±1.1%, whereas the localized percentage error was 0.5%±0.4% in LV myocardium and 0.5%±0.2% in LV blood pool. The localized absolute percentage error calculated for attenuation corrected SPECT reconstruction images was 3.2%±1.5% in LV myocardium and 2.5%±1.3% in LV blood pool. Conclusions: We developed a 3D model using a deep CNN with GAN training to estimate attenuation maps for SPECT directly and solely from the emission data. Qualitative and quantitative analysis demonstrated that the proposed method is capable of generating realistic attenuation maps with high accuracy, though further investigation with other tracers are needed.

Cross-vender, Cross-Tracer, and Cross-Protocol Deep Transfer Learning for Attenuation Map Generation of Cardiac SPECT

Attenuation Map Generation with Cross-Vendor and Cross-Tracer Transfer Learning for Cardiac SPECT

Generating Attenuation Map for SPECT-only Systems Using Generative Adversarial Networks

Investigation of Direct and Indirect Approaches of Deep-Learning-Based Attenuation Correction for General Purpose and Dedicated Cardiac SPECT Scanners

CT-free Attenuation Correction for Dedicated Cardiac SPECT Using a 3D Dual Squeeze-and-excitation Residual Dense Network

Cross-domain Iterative Network for Simultaneous Denoising, Limited-angle Reconstruction, and Attenuation Correction of Low-dose Cardiac SPECT

A Transfer Learning Framework for Deep Learning-Based CT-to-Perfusion Mapping on Lung Cancer Patients

DeTransUnet: attenuation correction of gated cardiac images without structural information

Deep-learning-based attenuation map generation in kidney single photon emission computed tomography

Deep-learning-based estimation of attenuation map improves attenuation correction performance over direct attenuation estimation for myocardial perfusion SPECT

Generative adversarial network-based attenuation correction for 99mTc-TRODAT-1 brain SPECT

Post-reconstruction Attenuation Correction for SPECT Myocardium Perfusion Imaging Facilitated by Deep Learning-Based Attenuation Map Generation

Three-dimensional Convolutional Neural Networks for Simultaneous Dual-Tracer PET Imaging

Direct attenuation correction for 99mTc-3PRGD2 chest SPECT lung cancer images using deep learning

Deep learning approach using SPECT-to-PET translation for attenuation correction in CT-less myocardial perfusion SPECT imaging

Automatic attenuation map estimation from SPECT data only for brain perfusion scans using convolutional neural networks

Direct and Indirect Strategies of Deep-Learning-based Attenuation Correction for General Purpose and Dedicated Cardiac SPECT

Deep learning-based attenuation correction in the image domain for myocardial perfusion SPECT imaging

A physics and learning-based transmission-less attenuation compensation method for SPECT

Direct Reconstruction for Simultaneous Dual-Tracer PET Imaging Based on Multi-Task Learning

Super-resolution reconstruction for parallel-beam SPECT based on deep learning and transfer learning: a preliminary simulation study