Department of Electrical and Computer Engineering, College of Engineering, University of Iowa, Iowa City, Iowa, USA.
Department of Radiology, Carver College of Medicine, University of Iowa, Iowa City, Iowa, USA.
Med Phys. 2024 Jun;51(6):4258-4270. doi: 10.1002/mp.17003. Epub 2024 Feb 28.
Osteoporosis is a bone disease related to increased bone loss and fracture-risk. The variability in bone strength is partially explained by bone mineral density (BMD), and the remainder is contributed by bone microstructure. Recently, clinical CT has emerged as a viable option for in vivo bone microstructural imaging. Wide variations in spatial-resolution and other imaging features among different CT scanners add inconsistency to derived bone microstructural metrics, urging the need for harmonization of image data from different scanners.
This paper presents a new deep learning (DL) method for the harmonization of bone microstructural images derived from low- and high-resolution CT scanners and evaluates the method's performance at the levels of image data as well as derived microstructural metrics.
We generalized a three-dimensional (3D) version of GAN-CIRCLE that applies two generative adversarial networks (GANs) constrained by the identical, residual, and cycle learning ensemble (CIRCLE). Two GAN modules simultaneously learn to map low-resolution CT (LRCT) to high-resolution CT (HRCT) and vice versa. Twenty volunteers were recruited. LRCT and HRCT scans of the distal tibia of their left legs were acquired. Five-hundred pairs of LRCT and HRCT image blocks of voxels were sampled for each of the twelve volunteers and used for training in supervised as well as unsupervised setups. LRCT and HRCT images of the remaining eight volunteers were used for evaluation. LRCT blocks were sampled at 32 voxel intervals in each coordinate direction and predicted HRCT blocks were stitched to generate a predicted HRCT image.
Mean ± standard deviation of structural similarity (SSIM) values between predicted and true HRCT using both 3DGAN-CIRCLE-based supervised (0.84 ± 0.03) and unsupervised (0.83 ± 0.04) methods were significantly (p < 0.001) higher than the mean SSIM value between LRCT and true HRCT (0.75 ± 0.03). All Tb measures derived from predicted HRCT by the supervised 3DGAN-CIRCLE showed higher agreement (CCC [0.956 0.991]) with the reference values from true HRCT as compared to LRCT-derived values (CCC [0.732 0.989]). For all Tb measures, except Tb plate-width (CCC = 0.866), the unsupervised 3DGAN-CIRCLE showed high agreement (CCC [0.920 0.964]) with the true HRCT-derived reference measures. Moreover, Bland-Altman plots showed that supervised 3DGAN-CIRCLE predicted HRCT reduces bias and variability in residual values of different Tb measures as compared to LRCT and unsupervised 3DGAN-CIRCLE predicted HRCT. The supervised 3DGAN-CIRCLE method produced significantly improved performance (p < 0.001) for all Tb measures as compared to the two DL-based supervised methods available in the literature.
3DGAN-CIRCLE, trained in either unsupervised or supervised fashion, generates HRCT images with high structural similarity to the reference true HRCT images. The supervised 3DGAN-CIRCLE improves agreements of computed Tb microstructural measures with their reference values and outperforms the unsupervised 3DGAN-CIRCLE. 3DGAN-CIRCLE offers a viable DL solution to retrospectively improve image resolution, which may aid in data harmonization in multi-site longitudinal studies where scanner mismatch is unavoidable.
骨质疏松症是一种与骨量流失和骨折风险增加相关的骨骼疾病。骨骼强度的可变性部分由骨密度（BMD）解释，其余部分由骨骼微观结构贡献。最近，临床 CT 已成为一种可行的体内骨骼微观结构成像选择。不同 CT 扫描仪之间的空间分辨率和其他成像特征的广泛差异使衍生的骨骼微观结构指标不一致，迫切需要协调来自不同扫描仪的图像数据。
本文提出了一种新的深度学习（DL）方法，用于协调来自低分辨率 CT（LRCT）和高分辨率 CT（HRCT）扫描仪的骨骼微观结构图像，并评估该方法在图像数据和衍生微观结构指标两个层面的性能。
我们推广了一种三维（3D）版本的 GAN-CIRCLE，该版本应用了两个受相同、残差和循环学习集（CIRCLE）约束的生成对抗网络（GAN）。两个 GAN 模块同时学习将 LRCT 映射到 HRCT 和反之亦然。招募了 20 名志愿者。获取他们左腿远端胫骨的 LRCT 和 HRCT 扫描。对每个志愿者的 12 个样本采集了 500 对 的 LRCT 和 HRCT 图像块，用于监督和非监督设置的训练。对其余 8 名志愿者的 LRCT 和 HRCT 图像进行了评估。LRCT 块在每个坐标方向以 32 个体素的间隔进行采样，并预测 HRCT 块进行拼接，以生成预测的 HRCT 图像。
使用基于 3DGAN-CIRCLE 的监督（0.84 ± 0.03）和非监督（0.83 ± 0.04）方法，预测的 HRCT 与真实 HRCT 之间的结构相似性（SSIM）值的平均值 ± 标准偏差显著（p < 0.001）高于 LRCT 与真实 HRCT 之间的平均 SSIM 值（0.75 ± 0.03）。所有通过监督 3DGAN-CIRCLE 从预测的 HRCT 中得出的 Tb 测量值与参考值（CCC [0.956 0.991]）的一致性都高于 LRCT 衍生值（CCC [0.732 0.989]）。对于所有 Tb 测量值，除了 Tb 板宽度（CCC = 0.866）之外，基于无监督 3DGAN-CIRCLE 的方法与真实 HRCT 衍生的参考测量值具有高度一致性（CCC [0.920 0.964]）。此外，Bland-Altman 图表明，与 LRCT 和无监督 3DGAN-CIRCLE 预测的 HRCT 相比，监督 3DGAN-CIRCLE 预测的 HRCT 减少了不同 Tb 测量值的残余值的偏差和可变性。与文献中两种基于深度学习的监督方法相比，监督 3DGAN-CIRCLE 方法在所有 Tb 测量值方面均表现出显著提高的性能（p < 0.001）。
以监督或非监督方式训练的 3DGAN-CIRCLE 生成的 HRCT 图像与参考真实 HRCT 图像具有高度的结构相似性。监督 3DGAN-CIRCLE 提高了计算的 Tb 微观结构测量值与参考值的一致性，并优于无监督 3DGAN-CIRCLE。3DGAN-CIRCLE 为回顾性提高图像分辨率提供了一种可行的深度学习解决方案，这可能有助于在扫描仪不匹配不可避免的多站点纵向研究中进行数据协调。
