Method and system for generating attenuation map from SPECT emission data

1. A system for estimating attenuation coefficients or attenuation maps (ATTMAP) from only single photon emission computed tomography (SPECT) emission data using artificial neural networks, comprising:

a machine learning system based upon deep artificial neural networks for estimating attenuation maps for SPECT emission data consisting essentially of images reconstructed from a photopeak window and/or one or more scatter windows, without requiring additional computed tomography (CT) or other transmission images, the machine learning system both generates attenuation map images from the SPECT emission data, wherein images reconstructed from the photopeak window and/or the scatter window are concatenated/used as a multi-channel/single-channel image, and enforces output attenuation map images to be consistent with a ground truth attenuation map generated based upon empirical evidence.

2. The system according to claim 1 , wherein the machine learning system includes a generator network and a discriminator network, the generator network generates attenuation map images from the SPECT emission data, wherein images reconstructed from the photopeak window and/or the scatter window are concatenated/used as a multi-channel/single-channel image and fed into the generator network, and the discriminator enforces output attenuation map images of the generator network to be consistent with the ground truth attenuation map generated based upon empirical evidence.

3. The system according to claim 2 , wherein the generator network is trained.

4. The system according to claim 3 , wherein the generator network is trained with Generative Adversarial Network training.

5. The system according to claim 3 , wherein the generator network is trained with an Adam optimizer.

6. The system according to claim 2 , wherein the discriminator network is trained.

7. The system according to claim 6 , wherein the discriminator network is trained with an Adam optimizer.

8. The system according to claim 2 , wherein the generator network is a deep convolutional neural network.

9. The system according to claim 8 , wherein the discriminator network is a deep convolutional neural network.

10. The system according to claim 2 , wherein the discriminator network is a deep convolutional neural network.

11. A method for generating attenuation maps and performing associated attenuation correction from SPECT emission data consisting essentially of images reconstructed from a photopeak window and/or one or more scatter windows, without requiring additional computed tomography (CT) or other transmission images, comprising:

generating an attenuation map from a NAC SPECT image dataset, comprising images reconstructed from the photopeak window and/or the scatter window, through deep learning, wherein generating is performed using a machine learning system that both generates attenuation map images from the SPECT emission data and enforces output attenuation map images;

estimating attenuated projection data via forward projecting the NAC SPECT image without the attenuation map to create estimated attenuated projection data; and

reconstructing an AC SPECT image from the estimated attenuated projection data using iterative reconstruction with attenuation correction by incorporating the attenuation map generated by deep learning.

12. The method according to claim 11 , where the machine learning system is based upon artificial neural networks.

13. The method according to claim 12 , wherein the artificial neural network includes a generator network.

14. The method according to claim 13 , wherein the generator network is a deep convolutional neural network.