Physics Research Laboratory, Department of Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, California;
Biomedical Engineering, Yale University, New Haven, Connecticut.
J Nucl Med. 2021 Nov;62(11):1645-1652. doi: 10.2967/jnumed.120.256396. Epub 2021 Feb 26.
Dedicated cardiac SPECT scanners with cadmium-zinc-telluride cameras have shown capabilities for shortened scan times or reduced radiation doses, as well as improved image quality. Since most dedicated scanners do not have integrated CT, image quantification with attenuation correction (AC) is challenging and artifacts are routinely encountered in daily clinical practice. In this work, we demonstrated a direct AC technique using deep learning (DL) for myocardial perfusion imaging (MPI). In an institutional review board-approved retrospective study, 100 cardiac SPECT/CT datasets with Tc-tetrofosmin, obtained using a scanner specifically with a small field of view, were collected at the Yale New Haven Hospital. A convolutional neural network was used for generating DL-based attenuation-corrected SPECT (SPECT) directly from noncorrected SPECT (SPECT) without undergoing an additional image reconstruction step. The accuracy of SPECT was evaluated by voxelwise and segmentwise analyses against the reference, CT-based AC (SPECT), using the 17-segment myocardial model of the American Heart Association. Polar maps of representative (best, median, and worst) cases were visually compared to illustrate potential benefits and pitfalls of the DL approach. The voxelwise correlations with SPECT were 92.2% ± 3.7% (slope, 0.87; = 0.81) and 97.7% ± 1.8% (slope, 0.94; = 0.91) for SPECT and SPECT, respectively. The segmental errors of SPECT scattered from -35% to 21% ( < 0.001), whereas the errors of SPECT stayed mostly within ±10% ( < 0.001). The average segmental errors (mean ± SD) were -6.11% ± 8.06% and 0.49% ± 4.35% for SPECT and SPECT, respectively. The average absolute segmental errors were 7.96% ± 6.23% and 3.31% ± 2.87% for SPECT and SPECT, respectively. Review of polar maps revealed successful reduction of attenuation artifacts; however, the performance of SPECT was not consistent for all subjects, likely because of different amounts of attenuation and different uptake patterns. We demonstrated the feasibility of direct AC using DL for SPECT MPI. Overall, our DL approach reduced attenuation artifacts substantially compared with SPECT, justifying further studies to establish safety and consistency for clinical applications in stand-alone SPECT systems suffering from attenuation artifacts.
使用带有碲化镉锌探测器的专用心脏 SPECT 扫描仪可以缩短扫描时间或降低辐射剂量，并提高图像质量。由于大多数专用扫描仪没有集成 CT，因此使用衰减校正 (AC) 进行图像量化具有挑战性，并且在日常临床实践中经常会遇到伪影。在这项工作中，我们展示了一种使用深度学习 (DL) 进行心肌灌注成像 (MPI) 的直接 AC 技术。在一项机构审查委员会批准的回顾性研究中，从耶鲁纽黑文医院收集了 100 例使用具有小视野的专用扫描仪获得的 Tc-四氮茂 SPECT/CT 数据集。使用卷积神经网络从未经校正的 SPECT (SPECT) 直接生成基于 DL 的衰减校正 SPECT (SPECT)，而无需进行额外的图像重建步骤。使用美国心脏协会的 17 节心肌模型，通过体素和节段分析，将 SPECT 的准确性与参考 CT 基 AC (SPECT) 进行了评估。使用代表性（最佳、中位数和最差）病例的极坐标图进行视觉比较，以说明 DL 方法的潜在优势和缺陷。与 SPECT 的体素相关性分别为 92.2%±3.7%（斜率，0.87； = 0.81）和 97.7%±1.8%（斜率，0.94； = 0.91）。SPECT 的节段性误差从-35%到 21%（<0.001），而 SPECT 的误差大多在±10%（<0.001）内。SPECT 的平均节段性误差（均值±SD）分别为-6.11%±8.06%和 0.49%±4.35%。SPECT 的平均绝对节段性误差分别为 7.96%±6.23%和 3.31%±2.87%。极坐标图的回顾表明衰减伪影的成功减少；然而，SPECT 的性能并不适用于所有患者，这可能是由于衰减量和摄取模式不同所致。我们证明了使用 DL 对 SPECT MPI 进行直接 AC 的可行性。总体而言，与 SPECT 相比，我们的 DL 方法大大减少了衰减伪影，这证明了在因衰减伪影而受到影响的独立 SPECT 系统中进行安全性和一致性的进一步研究是合理的。
