Subjects
One hundred and two subjects with suspected bone metastases from March 2021 to March 2022 were enrolled in this prospective study. All patients signed informed consent before examination. This study was approved by the institutional review board of Shanghai East Hospital. The subjects with a history of renal insufficiency, hormone, endocrine therapy, chemotherapy, and other treatments affecting bone metabolism were excluded from the research. The subject information was obtained from the medical record.
Spect/ct Acquisition
All patients were injected with 19–22 MBq/kg 99mTc-MDP. The whole-body scan and quantitative bone SPECT/CT imaging were performed using SPECT/CT (Siemens Symbia Intevo, Erlangen, Germany). Whole-body planar imaging was performed at about 3–4 hours post-injection. The areas with suspected malignancy were examined with a 20 minutes SPECT/CT (standard time) and followed by a 3 minutes SPECT (7 times reduction in scan time), the patient was instructed to remain still during the examination. Cases with motion artifacts were discarded after examination. The scanning matrix was 256×256, and the zoom factor was 1.0. Step-and-shoot mode with a total of 120 projections (60 steps) over 360° was used while 20s per step was adopted for standard time SPECT and 3s per step was for 1/7 standard time SPECT. Subsequently, a low-dose CT scan was performed at 130 kV and 10 mAs. CT data was reconstructed using a sharp bone kernel with 5mm slice thickness (B50s) and a smooth attenuation-correction kernel with 3mm slice thickness (B31s). SPECT reconstruction with attenuation correction was performed using the B31s CT attenuation map. The ordered subsets conjugate gradient enhanced xSPECT reconstruction algorithm (xSPECT/CT, Siemens Symbia Intevo) with 2 subsets and 28 iterations without post-smoothing was used to generate quantification measurement such as the maximum standard uptake value (SUVmax).
Imaging Process
A pretrained deep learning model[18] with integrated multi-scale and multi-modality features from 3min SPECT image and corresponding CT image was applied to generated enhanced SPECT images (3min-DL SPECT). The archiecture of the network was shown in Figure. 1. 3min SPECT SUV images were directly used as the network input without extra scaling. CT images were interpolated to the size of 3min SPECT and were normalized by image mean before network input. 3 consecutive SPECT and CT images were fed each time and average value was calculated if one slice was inferred multiple times.
Image Evaluation
20min, 3min and 3min-DL SPECT/CT images were evaluated independently by 2 nuclear medicine physicians with 5 and 10 years’ experience respectively. 5-point Likert scale (1, unacceptable image quality; 2, suboptimal image quality; 3, acceptable image quality; 4, good image quality; 5, excellent image quality) was used to score the three groups of images to evaluate their overall image quality, 99mTc-MDP details, presence of artifacts, and general diagnostic confidence. A score of 3 or higher indicate the requirement on image quality for clinical diagnosis was met. The lesion with the highest SUV in each subject was defined as the volume of interest (VOI), which was drawn using Siemens 3D Isocontour with SUVmax automatically calculated. All analyses were performed blind to the image acquisition information.
To quantitatively evaluate the performance of synthesized images, PSNR and SSIM are used as evaluation metrics. 20 min SPECT/CT were treated as the ground truth images. PSNR for synthesized image is defined as
$$\text{P}\text{S}\text{N}\text{R}=10\cdot {\text{log}}_{10}\left(\frac{\text{M}\text{A}{\text{X}}_{\text{g}\text{t}}^{2}}{\text{M}\text{S}\text{E}}\right)$$
Where \(\text{M}\text{A}{\text{X}}_{\text{g}\text{t}}\) is the maximum pixel value of ground truth 20min SPECT. MSE is the mean square error of synthesized images compared to the 20min SPECT.
SSIM for synthesized image is defined as
$$\text{S}\text{S}\text{I}\text{M}\left(\text{x},\text{y}\right)=\frac{\left(2{{\mu }}_{\text{x}}{{\mu }}_{\text{y}}+{\text{c}}_{1}\right)\left(2{{\sigma }}_{\text{x}\text{y}}+{\text{c}}_{2}\right)}{\left({{\mu }}_{\text{x}}^{2}+{{\mu }}_{\text{y}}^{2}+{\text{c}}_{1}\right)\left({{\sigma }}_{\text{x}}^{2}+{{\sigma }}_{\text{y}}^{2}+{\text{c}}_{2}\right)}$$
Where \({{\mu }}_{\text{x}}\) and\({ {\sigma }}_{\text{x}}^{2}\) are average value and variance of input synthesized image. \({{\mu }}_{\text{y}}\) and \({{\sigma }}_{\text{y}}^{2}\) are the average value and variance of input 20min SPECT. \({{\sigma }}_{\text{x}\text{y}}\) is the covariance of the two images. \({\text{c}}_{1}\)and \({\text{c}}_{2}\) are small constants. SSIM is calculated using the scikit-image package.
Diagnostic Performance
20min SPECT/CT images of all cases were read by the two reviewers. Benign and malignant lesions were determined based on pathological diagnosis, imaging examinations (20min SPECT/CT, CT, MRI), and clinical follow-up data. If the results were inconsistent, another senior physician was added to make a judgment, and the final benign and malignant lesions were determined as the gold standard. To dilute the memory effect, two readers independently read the disordered 3min and 3min-DL SPECT after one month. Benign (negative) and malignant (positive) were determined (a 5-point Likert scale of 1 was negative). Sensitivity, specificity, accuracy, and interobserver agreement were calculated.
Statistical analysis
Statistical analyses were performed using Graphpad Prism (8.0.0). Kappa consistency test was used to evaluate the consistency of the two reviewers in 20min, 3min, and 3min-DL SPECT images. Kappa ≥ 0.75 showed good consistency, 0.4 < Kappa < 0.75, moderate consistency, and Kappa ≤ 0.4, poor consistency. PSNR and RMSE in the quantitative analysis were compared with the paired Student’s t-test. Pearson and linear regression were used for demonstrating the consistency of the SUVmax between the 3min, 3min-DL, and 20min SPECT/CT. Chi-square test or Fisher's exact test was assessed the differences in sensitivity, specificity, accuracy. P < 0.05 was considered statistically significant.