Search

US-20260127798-A1 - CT-FREE ATTENUATION CORRECTION FOR SPECT USING DEEP LEARNING WITH IMAGING AND NON-IMAGING INFORMATION

US20260127798A1US 20260127798 A1US20260127798 A1US 20260127798A1US-20260127798-A1

Abstract

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.

Inventors

  • Chi Liu
  • Bo Zhou
  • Xiongchao Chen

Assignees

  • YALE UNIVERSITY

Dates

Publication Date
20260507
Application Date
20251114

Claims (20)

  1. 1 - 44 . (canceled)
  2. 45 . A system based upon artificial neural networks to generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography) data, comprising: a SPECT scanner; and 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 or generating an intermediate step of attenuation map estimation, wherein images from photopeak window with or without 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.
  3. 46 . The system according to claim 45 , wherein 126 keV-155 keV is used for the photopeak window.
  4. 47 . The system according to claim 45 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network includes 3D Dual Residual Dense Blocks.
  5. 48 . The system according to claim 45 , wherein each 3D Dual Residual Dense Block consists of a Residual Dense Block and a 3D Dual Squeeze-and-Excitation Block.
  6. 49 . The system according to claim 48 , wherein the 3D Dual Squeeze-and-Excitation Block consists of two squeeze-and-excitation branches.
  7. 50 . The system according to claim 49 , wherein the squeeze-and-excitation branches include a spatial-squeeze-and-channel-excitation for re-calibrating feature channels and a channel-squeeze-and-spatial-excitation for recalibrating spatial features.
  8. 51 . The system according to claim 48 , wherein the Residual Dense Block ensures that each convolutional layer in the Residual Dense Block has access to all the subsequent layers and passes on information that needs to be preserved.
  9. 52 . The system according to claim 51 , wherein the Residual Dense Block includes multiple convolutional layers with Rectified Linear Units and a local feature fusion.
  10. 53 . The system according to claim 45 , 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 Dual Squeeze-and-Excitation Block.
  11. 54 . A method based upon artificial neural networks to directly generate attenuation-corrected SPECT from non-attenuation-corrected SPECT (single photon emission computed tomography), comprising: generating images from a photopeak window with or without one or more scatter windows of a SPECT scanner, and 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 or generating an intermediate step of attenuation map estimation, wherein images from photopeak window with or without 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.
  12. 55 . The method according to claim 54 , wherein 126 keV-155 keV is used for the photopeak window.
  13. 56 . The method according to claim 54 , wherein the 3D Dual Squeeze-and-Excitation Residual Dense Network includes 3D Dual Residual Dense Blocks.
  14. 57 . The method according to claim 56 , wherein each 3D Dual Residual Dense Block consists of a Residual Dense Block and a 3D Dual Squeeze-and-Excitation Block.
  15. 58 . The method according to claim 57 , wherein the 3D Dual Squeeze-and-Excitation Block consists of two squeeze-and-excitation branches.
  16. 59 . The method according to claim 58 , wherein the squeeze-and-excitation branches include a spatial-squeeze-and-channel-excitation for re-calibrating feature channels and a channel-squeeze-and-spatial-excitation for recalibrating spatial features.
  17. 60 . The method according to claim 57 , wherein the Residual Dense Block ensures that each convolutional layer in the Residual Dense Block has access to all the subsequent layers and passes on information that needs to be preserved.
  18. 61 . The method according to claim 60 , wherein the Residual Dense Block includes multiple convolutional layers with Rectified Linear Units and a local feature fusion.
  19. 62 . The method according to claim 55 , 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.
  20. 63 . The method according to claim 55 , 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.

Description

This invention was made with government support under R01HL123949 and R01HL154345 awarded by National Institutes of Health. The government has certain rights in the invention. BACKGROUND OF THE INVENTION 1. Field of the Invention The invention relates to attenuation correction for single photon emission computed tomography (SPECT) through the application of deep learning networks incorporating images from scatter window(s) and individual subject's information such as BMI (body mass index) and gender. 2. Description of the Related Art Single photon emission computed tomography (SPECT) is a non-invasive imaging procedure that provides radiotracer distribution images of the patient body by detecting gamma-ray photons. SPECT plays an important role in the clinical diagnosis of cardiovascular, oncological, and neurological disease. In order to perform qualitative, quantitative, or semi-quantitative analysis for SPECT, accurate attenuation correction is essential. SPECT continues to play a critical role in the diagnosis and management of coronary artery disease (CAD). While conventional SPECT scanners using parallel-hole collimators are still the foundation of cardiac SPECT, dedicated cardiac SPECT scanners have also been developed. Such dedicated scanners, such as the GE Alcyone 530/570c systems and the D-SPECT™ systems (that is, a medical imaging apparatus featuring nuclear imaging, namely, SPECT imaging as manufactured by Spectrum Dynamics Medical Limited Company), both with CZT (Cadmium Zinc Telluride) detectors, typically have multiple detectors collecting photons emitted from the heart simultaneously. This leads to dramatically improved sensitivity (for example, 2 to 5 times). In addition, the GE Alcyone 530/570c systems use pinhole collimators and can achieve much higher resolution. These dedicated scanners have opened doors to new applications with significant clinical impact, including, but not limited to, ultra-low-dose imaging, absolute quantification of myocardial blood flow (MBF) and coronary flow reserve (CFR), high resolution molecular imaging, multi-isotope imaging, and motion correction. Most of these new applications are uniquely achievable only using dedicated scanners. However, an artifact-free reconstruction of radiotracer distribution and absolute activity for SPECT can only be obtained with the assistance of accurate correction of photon attenuation using an individualized attenuation map [Quantitative analysis in nuclear medicine imaging. Springer (2006)]. Therefore, in current clinical practice, computed tomography (CT) is utilized to generate the attenuation map [Pan, T. S., King, M. A., Luo, D. S., Dahlberg, S. T., Villegas, B. J.: Estimation of attenuation maps from scatter and photopeak window single photon-emission computed tomographic images of technetium 99m-labeled sestamibi. Journal of Nuclear Cardiology 4 (1) (1997) 42-51; Zaidi, H., Hasegawa, B.: Determination of the attenuation map in emission tomography. Journal of Nuclear Medicine 44 (2) (2003) 291-315; Pan, T. S., King, M. A., De Vries, D. J., Ljungberg, M.: Segmentation of the body and lungs from compton scatter and photopeak window data in spect: a monte-carlo investigation. IEEE transactions on medical imaging 15 (1) (1996) 13-24]. However, notably, about 80% of SPECT scanners are stand-alone scanners and images are reconstructed without CT transmission scanning. There are two additional limitations even when the CT is available. First, the use of adjunctive CT scanning for SPECT attenuation introduces additional radiation to patients along with increasing imaging system cost. Secondly, the misalignment between CT and SPECT due to motion can cause attenuation correction artifacts leading to inaccurate assessment of regional myocardial activity [Schäfers, K. P., Stegger, L.: Combined imaging of molecular function and morphology with pet/ct and spect/ct: image fusion and motion correction. Basic research in cardiology 103 (2) (2008) 191-199; McQuaid, S. J., Hutton, B. F.: Sources of attenuation-correction artefacts in cardiac pet/ct and spect/ct. European journal of nuclear medicine and molecular imaging 35 (6) (2008) 1117-1123]. Previous works have attempted to estimate the attenuation map directly from SPECT emission data [Jha, A. K., Zhu, Y., Clarkson, E., Kupinski, M. A., Frey, E. C.: Fisher information analysis of list-mode spect emission data for joint estimation of activity and attenuation distribution. arXiv preprint arXiv:1807.01767 (2018); Cade, S. C., Arridge, S., Evans, M. J., Hutton, B. F.: Use of measured scatter data for the attenuation correction of single photon emission tomography without trans-mission scanning. Medical physics 40 (8) (2013) 082506]. Unfortunately, attenuation map estimation from these methods involving iterative optimization is time-consuming, and often contain high noise level when image activity is relatively low, which can result in SPECT reconstruction errors. Recently, a deep-learni