Voxel-based correction of CT attenuations for accurate quantification of coronary artery calcification in low tube voltage scans with deep learning reconstruction.
Authors
Affiliations (12)
Affiliations (12)
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- CT Imaging Research Center, GE HealthCare China, Shanghai, China. Electronic address: [email protected].
- CT Imaging Research Center, GE HealthCare China, Shanghai, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China; School of Medical Imaging, Xuzhou Medical University, Xuzhou, Jiangsu, China; Jiangsu Medical Imaging and Digital Medicine Engineering Research Center, Xuzhou, Jiangsu, Province, China. Electronic address: [email protected].
Abstract
Low-kilovolt (kV), low-dose scanning combined with deep learning-based image reconstruction (DLIR) is increasingly adopted in clinical practice. However, it often introduces biases in the quantification and risk stratification of coronary artery calcification (CAC). This study proposes a voxel-based CT attenuation correction method to enable accurate CAC assessment under low-kV imaging. Phantom scans containing various inserts were acquired using a standard protocol (120 kVp with filtered back projection, STD group) and a low-kV protocol (80 kVp with DLIR, low-kV group). A linear regression model was established to derive a correction formula mapping CT attenuation from the low-kV to the STD. Subsequently, patients referred for CAC scoring were prospectively enrolled. Each patient underwent two scans (STD and low-kV). Voxel-wise CT attenuations in the low-kV images were corrected using the phantom-derived calibration formula. Automated CAC analysis software was used to compute calcified volume, equivalent mass, and Agatston score, followed by risk stratification into standard categories (0, 10, 100, 400). Corrected low-kV measurements were compared to those from the STD. Objective image quality was assessed through CT attenuation, standard deviation (SD) and signal-to-noise ratio (SNR). Subjective quality was evaluated using a 5-point Likert scale. A total of 190 patients were included. The low-kV group achieved a 77.6 % reduction in radiation dose compared to the STD group. Prior to correction, the low-kV group significantly overestimated calcified volume, equivalent mass, Agatston score, and risk category (all P < 0.05). After voxel-based correction, no statistically significant differences remained compared to the STD group (all P > 0.05). The bias in calcified volume, equivalent mass, and Agatston score were reduced from 48.14 ± 73.66, 19.48 ± 33.44, and 62.44 ± 94.46 to 6.63 ± 23.56, -0.44 ± 6.68, and 3.05 ± 28.25, respectively. The risk stratification misclassification rate decreased from 20.53 % to 5.79 %. The low-kV group outperformed the STD group in objective image assessments, showing superior CT attenuation, SD and SNR. There were no significant differences in subjective image assessments. The proposed voxel-based correction method effectively mitigates the overestimation bias introduced by low-kV protocols in CAC assessment.