CT-free attenuation correction for SPECT using deep learning with imaging and non-imaging information
A system based upon artificial neural networks generates attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) without or with an intermediate step of attenuation map estimation. The system includes a SPECT scanner with CZT cameras for dynamic SPECT imaging. The system also includes a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for generating attenuation-corrected SPECT or attenuation maps from non-attenuation-corrected SPECT. The machine learning system reconstructs images from photopeak window and one or more scatter windows of the SPECT scanner are fed to the 3D Dual Squeeze-and-Excitation Residual Dense Network to generate attenuation-corrected SPECT or attenuation maps.
1 . A system based upon artificial neural networks to directly generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) without any intermediate step of attenuation map estimation, comprising:
a SPECT scanner for dynamic SPECT imaging;
a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for directly generating attenuation-corrected SPECT from non-attenuation-corrected SPECT without any intermediate step of attenuation map estimation, wherein the machine learning system reconstructs images from a photopeak window image of the SPECT scanner, and the reconstructed images are fed to the 3D Dual Squeeze-and-Excitation Residual Dense Network to generate attenuation-corrected SPECT.
2 . The system according to claim 1 , wherein 126 keV-155 keV is used for the photopeak window.
3 . A system based upon artificial neural networks to directly generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) without any intermediate step of attenuation map estimation, comprising:
a SPECT scanner for dynamic SPECT imaging;
a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for directly generating attenuation-corrected SPECT from non-attenuation-corrected SPECT without any intermediate step of attenuation map estimation, wherein the machine learning system reconstructs images from a photopeak window image and one or more scatter window images of the SPECT scanner, and the reconstructed images are fed to the 3D Dual Squeeze-and-Excitation Residual Dense Network to generate attenuation-corrected SPECT;
wherein 126 keV-155 keV is used for the photopeak window and 55-80 keV, 80-105 keV, and 105-130 keV are used for the scatter windows.
4 . The system according to claim 1 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network includes 3D Dual Residual Dense Blocks.
5 . The system according to claim 1 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network consists of a Residual Dense Block, a Dual Squeeze-and-Excitation block, and a U-Net backbone architecture supported by the Residual Dense Block and the 3D Dual Squeeze-and-Excitation Block.
6 . A method based upon artificial neural networks to directly generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) without any intermediate step of attenuation map estimation, comprising:
generating images from a photopeak window and one or more scatter windows of a SPECT scanner;
applying a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for directly generating attenuation-corrected SPECT from non-attenuation-corrected SPECT without any intermediate step of attenuation map estimation, wherein the machine learning system reconstructs the images from the photopeak window image of the SPECT scanner, and the reconstructed images are to generate attenuation-corrected SPECT.
7 . The method according to claim 6 , wherein 126 keV-155 keV is used for the photopeak window.
8 . A method based upon artificial neural networks to directly generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) without any intermediate step of attenuation map estimation, comprising:
generating images from a photopeak window and one or more scatter windows of a SPECT scanner;
applying a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for directly generating attenuation-corrected SPECT from non-attenuation-corrected SPECT without any intermediate step of attenuation map estimation, wherein the machine learning system reconstructs the images from the photopeak window image and the one or more scatter window images of the SPECT scanner, and the reconstructed images are used to generate attenuation-corrected SPECT;
wherein 126 keV-155 keV is used for the photopeak window and 55-80 keV, 80-105 keV, and 105-130 keV are used for the scatter windows.
9 . The method according to claim 6 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network includes 3D Dual Residual Dense Blocks.
10 . The method according to claim 6 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network consists of a Residual Dense Block, a Dual Squeeze-and-Excitation block, and a U-Net backbone architecture supported by the Residual Dense Block and 3D Dual Squeeze-and-Excitation Block.
11 . The method according to claim 6 , wherein the machine learning system reconstructs the images from the photopeak window image and the one or more scatter windows of a dedicated cardiac SPECT scanner having a small field of view of 19 cm in diameter, and further including the step of estimating truncated or full attenuation maps from SPECT reconstructions in the photopeak window and the one or more scatter windows of the SPECT scanners.
12 . A system based upon artificial neural networks to generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) with an intermediate step of attenuation map estimation, comprising:
a SPECT scanner for dynamic SPECT imaging;
a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for generating attenuation maps from non-attenuation-corrected SPECT, wherein the machine learning system reconstructs images from a photopeak window image of the SPECT scanner, and the reconstructed images are fed to the 3D Dual Squeeze-and-Excitation Residual Dense Network to generate an attenuation map.
13 . The system according to claim 12 , wherein 126 keV-155 keV is used for the photopeak window.
14 . A system based upon artificial neural networks to generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) with an intermediate step of attenuation map estimation, comprising:
a SPECT scanner for dynamic SPECT imaging;
a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for generating attenuation maps from non-attenuation-corrected SPECT, wherein the machine learning system reconstructs images from a photopeak window and one or more scatter windows of the SPECT scanner, and the reconstructed images are fed to the 3D Dual Squeeze-and-Excitation Residual Dense Network to generate attenuation-corrected SPECT;
wherein 126 keV-155 keV is used for the photopeak window and 55-80 keV, 80-105 keV, and 105-130 keV are used for the scatter windows.
15 . The system according to claim 12 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network includes 3D Dual Residual Dense Blocks.
16 . The system according to claim 12 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network consists of a Residual Dense Block, a Dual Squeeze-and-Excitation block, and a U-Net backbone architecture supported by the Residual Dense Block and the 3D Dual Squeeze-and-Excitation Block.
17 . A method based upon artificial neural networks to generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) with an intermediate step of attenuation map estimation, comprising:
generating images from a photopeak window and one or more scatter windows of a SPECT scanner;
applying a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for generating attenuation maps from non-attenuation-corrected SPECT, wherein the machine learning system reconstructs the images from the photopeak window image, and the reconstructed images are used to generate attenuation-corrected SPECT.
18 . The method according to claim 17 , wherein 126 keV-155 keV is used for the photopeak window.
19 . A method based upon artificial neural networks to generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) with an intermediate step of attenuation map estimation, comprising:
generating images from a photopeak window and one or more scatter windows of a SPECT scanner;
applying a machine learning system including a 3D Dual Squeeze-and-Excitation Residual Dense Network for generating attenuation maps from non-attenuation-corrected SPECT, wherein the machine learning system reconstructs the images from the photopeak window image and the one or more scatter window images of the SPECT scanner, and the reconstructed images are used to generate attenuation-corrected SPECT;
wherein 126 keV-155 keV is used for the photopeak window and 55-80 keV, 80-105 keV, and 105-130 keV are used for the scatter windows.
20 . The method according to claim 17 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network includes 3D Dual Residual Dense Blocks.
21 . The method according to claim 17 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network consists of a Residual Dense Block, a Dual Squeeze-and-Excitation block, and a U-Net backbone architecture supported by the Residual Dense Block and 3D Dual Squeeze-and-Excitation Block.
22 . The method according to claim 17 , wherein the machine learning system reconstructs the images from the photopeak window image and the one or more scatter windows of a dedicated cardiac SPECT scanner having a small field of view of 19 cm in diameter, and further including the step of estimating truncated or full attenuation maps from SPECT reconstructions in the photopeak window and the one or more scatter windows of the SPECT scanners.