CT-Less Attenuation Correction in Image Space Using Deep Learningfor Dedicated Cardiac SPECT: A Feasibility Study

Introduction: Dedicated cardiac SPECT scanners (e.g., GE Discovery NM 530c) with cadmium-zinc-telluride (CZT) cameras have shown capabilities of shortened scan times or reduced radiation doses as well as improved image quality. Since most of the dedicated scanners do not have an integrated CT, image quantification with attenuation correction (AC) is challenging and artifacts are routinely encountered in daily clinical practice. In this work, we demonstrate a CT-less attenuation correction technique using deep learning (DL) for dedicated cardiac SPECT.

Methods: In an IRB-approved study, 100 cardiac SPECT/CT datasets with ^99mTc-Tetrofosmin using a GE Discovery NM 570c SPECT/CT scanner were retrospectively collected at the Yale New Haven Hospital. A U-Net-based network was used for generating attenuation-corrected SPECT (SPECT_DL) directly from non-corrected SPECT (SPECT_NC) without undergoing an additional image reconstruction step. Quality of the DL-generated images (SPECT_DL) was evaluated by the normalized root mean square error (NRMSE), peak signal to noise ratio (PSNR), and structural similarity index (SSIM). Statistical analyses were performed using joint and error histograms, demonstrating voxel-wise correlation between SPECT_DL and SPECT_CTAC (SPECT with CT-based AC). The clinical value of this DL-based CT-less AC was illustrated through comparison of polar bull’s eye plots.

Results: In comparison to reference SPECT_CTAC, NRMSEs were 0.232 ± 0.077 and 0.148 ± 0.095; PSNRs 31.3 ± 2.8 and 36.2 ± 4.1; SSIMs 0.984 ± 0.008 and 0.993 ± 0.006 for SPECT_NC and SPECT_DL, respectively. These results were consistent with the joint histograms that showed the voxel-wise correlations of 92.2% ± 3.7 (slope = 0.87; R² = 0.81) for SPECT_NC and 97.7% ± 1.8 (slope = 0.94; R² = 0.91) for SPECT_DL, which was consistent with the comparison of their error histograms. Review of polar bull’s eye plots revealed successful demonstration of reduced attenuation artifacts; however, the performance of SPECT_DL was not consistent for all subjects likely because of patient-specific acquisition geometries and uptake patterns.

Conclusions: This study demonstrated the feasibility of CT-less AC using deep learning for potential clinical applications in dedicated cardiac SPECT not combined with CT. Furthermore, the results showed that CT-less AC could provide similar quantification accuracy as compared to the conventional CTAC for dedicated cardiac SPECT imaging. Acknowledgements: The study was supported by the National Institutes of Health under grants R01HL135490, R01EB026331, R01HL123949, and American Heart Association award 18PRE33990138.

• Download figure
• Open in new tab
• Download powerpoint

• Download figure
• Open in new tab
• Download powerpoint