A CNN-based denoising method trained with images acquired with electron density phantoms for thin-sliced coronary artery calcium scans.

PURPOSE: This work proposed a convolutional neural network (CNN)-based method trained with images acquired with electron density phantoms to reduce quantum noise for coronary artery calcium (CAC) scans reconstructed with slice thickness less than 3 mm. METHODS: A DenseNet model was used to estimate quantum noise for CAC scans reconstructed with slice thickness of 0.5, 1.0 and 1.5 mm. Training data was acquired using electron density phantoms in three different sizes. The label images of the CNN model were real noise maps, while the input images of the CNN model were pseudo noise maps. Image denoising was conducted by subtracting the CNN output images from thin-sliced CAC scans. The efficacy of the proposed method was verified through both phantom study and patient study. RESULTS: By means of phantom study, the proposed method was proven effective in reducing quantum noise in CAC scans reconstructed with 1.5-mm slice thickness without causing significant texture change or variation in HU values. With regard to patient study, calcifications were more clear on the denoised CAC scans reconstructed with slice thickness of 0.5, 1.0 and 1.5 mm than on 3-mm slice images, while over-smooth changes were not observed in the denoised CAC scans reconstructed with 1.5-mm slice thickness. CONCLUSION: Our results demonstrated that the electron density phantoms can be used to generate training data for the proposed CNN-based denoising method to reduce quantum noise for CAC scans reconstructed with 1.5-mm slice thickness. Because anthropomorphic phantom is not a necessity, our method could make image denoising more practical in routine clinical practice.

Investigating the Feasibility of Using DenseNet to Improve Coronary Calcification Detection in CT.

OBJECTIVES: Interscan reproducibility of coronary artery calcium (CAC) scoring can be improved by using a smaller slice thickness but at the cost of higher image noise. This study aimed to investigate the feasibility of using densely connected convolutional network (DenseNet) to reduce the image noise in CAC scans reconstructed with slice thickness < 3 mm for improving coronary calcification detection in CT. METHODS: Phantom data acquired with QRM and CIRS phantoms were used for model training and testing, where the DenseNet model adopted in this work was a convolutional neural network (CNN) designed for super resolution recovery. After phantom study, the proposed method was evaluated in terms of its ability to improve calcification detection using patient data. The CNN input images (IMGinput) were CAC scans reconstructed with 0.5-, 1.0- and 1.5-mm slice thickness, while CNN label images were CAC scans reconstructed with 3-mm slice thickness (IMG3mm). Region of interest (ROI) analysis was carried out on IMG3mm, IMGinput and CNN output images (IMGoutput). Two-sample t test was used to compare the difference in Hounsfield Unit (HU) values within ROI between IMG3mm and IMGoutput. RESULTS: For the calcifications in QRM phantoms, no statistically significant difference was found when comparing the HU values of 400- and 800-HA calcifications identified on IMG3mm to those on IMGoutput with slice thickness of 0.5, 1.0 or 1.5 mm. On the other hand, statistically significant difference was found when comparing the HU values of 200-HA calcifications identified on IMG3mm to those on IMGoutput with a slice thickness of 0.5 and 1.0 mm. Meanwhile, no statistically significant difference was found when comparing the HU values of 200-HA calcifications identified on IMG3mm to those on IMGoutput with a slice thickness of 1.5 mm. As for the rod inserts in CIRS phantoms simulating 9 different tissue types in human body, there was no statistically significant difference between IMG3mm and IMGoutput with slice thickness of 1.5 mm, and all the p values were larger than 0.10. With regards to patient study, more calcification pixels were detected on IMGoutput with a slice thickness of 1.5 mm than on IMG3mm, so calcifications were more clear on the denoised images. CONCLUSION: According to our results, the CNN-based denoising method could reduce statistical noise in IMGinput with a slice thickness of 1.5 mm without causing significant texture change or variation in HU values. The proposed method could improve cardiovascular risk prediction by detecting small and soft calcifications that are barely identified on 3-mm slice images used in conventional CAC scans.

Radiomics-Based Deep Learning Prediction of Overall Survival in Non-Small-Cell Lung Cancer Using Contrast-Enhanced Computed Tomography.

Patient outcomes of non-small-cell lung cancer (NSCLC) vary because of tumor heterogeneity and treatment strategies. This study aimed to construct a deep learning model combining both radiomic and clinical features to predict the overall survival of patients with NSCLC. To improve the reliability of the proposed model, radiomic analysis complying with the Image Biomarker Standardization Initiative and the compensation approach to integrate multicenter datasets were performed on contrast-enhanced computed tomography (CECT) images. Pretreatment CECT images and the clinical data of 492 patients with NSCLC from two hospitals were collected. The deep neural network architecture, DeepSurv, with the input of radiomic and clinical features was employed. The performance of survival prediction model was assessed using the C-index and area under the curve (AUC) 8, 12, and 24 months after diagnosis. The performance of survival prediction that combined eight radiomic features and five clinical features outperformed that solely based on radiomic or clinical features. The C-index values of the combined model achieved 0.74, 0.75, and 0.75, respectively, and AUC values of 0.76, 0.74, and 0.73, respectively, 8, 12, and 24 months after diagnosis. In conclusion, combining the traits of pretreatment CECT images, lesion characteristics, and treatment strategies could effectively predict the survival of patients with NSCLC using a deep learning model.

Applying MRI Intensity Normalization on Non-Bone Tissues to Facilitate Pseudo-CT Synthesis from MRI.

This study aimed to facilitate pseudo-CT synthesis from MRI by normalizing MRI intensity of the same tissue type to a similar intensity level. MRI intensity normalization was conducted through dividing MRI by a shading map, which is a smoothed ratio image between MRI and a three-intensity mask. Regarding pseudo-CT synthesis from MRI, a conversion model based on a three-layer convolutional neural network was trained and validated. Before MRI intensity normalization, the mean value ± standard deviation of fat tissue in 0.35 T chest MRI was 297 ± 73 (coefficient of variation (CV) = 24.58%), which was 533 ± 91 (CV = 17.07%) in 1.5 T abdominal MRI. The corresponding results were 149 ± 32 (CV = 21.48%) and 148 ± 28 (CV = 18.92%) after intensity normalization. With regards to pseudo-CT synthesis from MRI, the differences in mean values between pseudo-CT and real CT were 3, 15, and 12 HU for soft tissue, fat, and lung/air in 0.35 T chest imaging, respectively, while the corresponding results were 3, 14, and 15 HU in 1.5 T abdominal imaging. Overall, the proposed workflow is reliable in pseudo-CT synthesis from MRI and is more practicable in clinical routine practice compared with deep learning methods, which demand a high level of resources for building a conversion model.

CMR-derived ECVs vary with myocardial region and associate with the regional wall thickness.

This study was designed to identify whether the position and size of the region of interest (ROI) influence extracellular volume fraction (ECV) measurements. Patients with localized (n = 203) or infiltrative (n = 215) cardiomyopathies and 36 normal controls were enrolled in this study. ECV measurements at 4 different regions, including the anterior, septal, posterior and lateral wall regions, were measured, and all groups were compared. Regional ECV was correlated with the corresponding regional wall thickness. The diagnostic power to differentiate the myocardial abnormalities was evaluated for each myocardial region. ECVs measured using five different ROI sizes within each myocardial region were compared. Our results showed that ECVs varied among the myocardial regions, and this variation was significantly associated with regional wall thicknesses. For the detection of myocardial abnormalities, regional ECV revealed similar results as ECV derived from the whole region except for the anterior region. No significant difference was found in the ECVs measured using the five different ROI sizes. In conclusion, CMR-derived ECVs vary with myocardial region, and this variation is significantly associated with the regional wall thickness. In contrast, the measured size of the ROI has less of an effect on the ECV.

Optimization of HU threshold for coronary artery calcium scans reconstructed at 0.5-mm slice thickness using iterative reconstruction.

PURPOSE: This work investigated the simultaneous influence of tube voltage, tube current, body size, and HU threshold on calcium scoring reconstructed at 0.5-mm slice thickness using iterative reconstruction (IR) through multivariate analysis. Regression results were used to optimize the HU threshold to calibrate the resulting Agatston scores to be consistent with those obtained from the conventional protocol. METHODS: A thorax phantom set simulating three different body sizes was used in this study. A total of 14 coronary artery calcium (CAC) protocols were studied, including 1 conventional protocol reconstructed at 3-mm slice thickness, 1 FBP protocol, and 12 statistical IR protocols (3 kVp values*4 SD values) reconstructed at 0.5-mm slice thickness. Three HU thresholds were applied for calcium identification, including 130, 150, and 170 HU. A multiple linear regression method was used to analyze the impact of kVp, SD, body size, and HU threshold on the Agatston scores of three calcification densities for IR-reconstructed CAC scans acquired with 0.5-mm slice thickness. RESULTS: Each regression relationship has R2 larger than 0.80, indicating a good fit to the data. Based on the regression models, the HU thresholds as a function of SD estimated to ensure the quantification accuracy of calcium scores for 120-, 100-, and 80-kVp CAC scans reconstructed at 0.5-mm slice thickness using IR for three different body sizes were proposed. Our results indicate that the HU threshold should be adjusted according to the imaging condition, whereas a 130-HU threshold is appropriate for 120-kVp CAC scans acquired with SD = 55 for body size of 24.5 cm. CONCLUSION: The optimized HU thresholds were proposed for CAC scans reconstructed at 0.5-mm slice thickness using IR. Our study results may provide a potential strategy to improve the reliability of calcium scoring by reducing partial volume effect while keeping radiation dose as low as reasonably achievable.