Three-stage segmentation of lung region from CT images using deep neural networks.

BACKGROUND: Lung region segmentation is an important stage of automated image-based approaches for the diagnosis of respiratory diseases. Manual methods executed by experts are considered the gold standard, but it is time consuming and the accuracy is dependent on radiologists' experience. Automated methods are relatively fast and reproducible with potential to facilitate physician interpretation of images. However, these benefits are possible only after overcoming several challenges. The traditional methods that are formulated as a three-stage segmentation demonstrate promising results on normal CT data but perform poorly in the presence of pathological features and variations in image quality attributes. The implementation of deep learning methods that can demonstrate superior performance over traditional methods is dependent on the quantity, quality, cost and the time it takes to generate training data. Thus, efficient and clinically relevant automated segmentation method is desired for the diagnosis of respiratory diseases. METHODS: We implement each of the three stages of traditional methods using deep learning methods trained on five different configurations of training data with ground truths obtained from the 3D Image Reconstruction for Comparison of Algorithm Database (3DIRCAD) and the Interstitial Lung Diseases (ILD) database. The data was augmented with the Lung Image Database Consortium (LIDC-IDRI) image collection and a realistic phantom. A convolutional neural network (CNN) at the preprocessing stage classifies the input into lung and none lung regions. The processing stage was implemented using a CNN-based U-net while the postprocessing stage utilize another U-net and CNN for contour refinement and filtering out false positives, respectively. RESULTS: The performance of the proposed method was evaluated on 1230 and 1100 CT slices from the 3DIRCAD and ILD databases. We investigate the performance of the proposed method on five configurations of training data and three configurations of the segmentation system; three-stage segmentation and three-stage segmentation without a CNN classifier and contrast enhancement, respectively. The Dice-score recorded by the proposed method range from 0.76 to 0.95. CONCLUSION: The clinical relevance and segmentation accuracy of deep learning models can improve though deep learning-based three-stage segmentation, image quality evaluation and enhancement as well as augmenting the training data with large volume of cheap and quality training data. We propose a new and novel deep learning-based method of contour refinement.

COMPARISON OF CONVENTIONAL HAND EXAMINATION ON SIX OPTIMISED DR SYSTEMS.

The purpose of this study was to investigate the challenges in comparing digital radiography (DR) systems from different vendors for various combinations of exposure factors in posterior-anterior hand radiographs. Image quality was evaluated for a range of tube voltages and tube current-time products using a technical contrast-detail (CDRAD) phantom and an anthropomorphic hand phantom. 900 technical CDRAD images were analysed providing quality figures of merit (IQFinv) and two experienced reporting radiographers using visual grading analysis (VGA) scored 108 anthropomorphic images. This study demonstrates the differences between the DR systems included. When compensating for variations in dose, Canon showed superior results for technical image quality and Fuji for visual image quality for a standard dose point at DR hand examination (ln(DAP) 1.1, 50 kV and 2.5 mAs).

CAN SCATTER CORRECTION SOFTWARE REPLACE A GRID IN DR PELVIC EXAMINATIONS?

The purpose was to examine if scatter correction software could replace a grid while maintaining image quality and reducing radiation dose for pelvic DR examinations. Grid images was produced with 70 kV and 16mAs. Anthropomorphic- and Contrast Detail RADiography (CDRAD) non-grid images were produced with 60 kV, 80 kV and 90 kV combined with five different mAs and scatter correction software. The anthropomorphic images were analyzed by absolute Visual Grading Analysis (VGA). The CDRAD images were analyzed using the CDRAD analysis software. The results showed a total of 54.6% non-grid images were evaluated as unsuitable for diagnostic use by the VGA. The CDRAD grid images showed that the IQF_inv values were significantly different (p = 0.0001) when compared to every group of non-grid images. Hereby, the conclusion stated that the scatter correction software did not compensate for the loss in image quality due to scattered radiation at the exposure levels included in a pelvic examination.

Image quality in coronary computed tomography angiography: influence of adaptive statistical iterative reconstruction at various radiation dose levels.

Background Computed tomography (CT) technology is rapidly evolving and software solution developed to optimize image quality and/or lower radiation dose. Purpose To investigate the influence of adaptive statistical iterative reconstruction (ASIR) at different radiation doses in coronary CT angiography (CCTA) in detailed image quality. Material and Methods A total of 160 CCTA were reconstructed as follows: 55 scans with filtered back projection (FBP) (650 mA), 51 scans (455 mA) with 30% ASIR (ASIR30), and 54 scans (295 mA) with 60% ASIR (ASIR60). For each reconstruction, subjective image quality was assessed by five independent certified cardiologists using a visual grading analysis (VGA) with five predefined image quality criteria consisting of a 5-point scale. Objective measures were contrast, noise, and contrast-to-noise ratio (CNR). Results The CTDIvol resulted in 10.3 mGy, 7.4 mGy, and 4.6 mGy for FBP, ASIR30, and ASIR60, respectively. Homogeneity of the left ventricular lumen was the sole aspect in which reconstruction algorithms differed with a decreasing effect for ASIR60 compared to FBP (estimated odds ratio [OR] = 0.49 [95% confidence interval (CI) = 0.32-0.76; P = 0.001]). Decreased sharpness and spatial- and low-contrast resolutions were observed when using ASIR instead of FBP, but differences were not statistically significant. Concerning objective measurements, noise increased significantly for ASIR30 (OR = 1.08; 95% CI = 1.02-1.14; P = 0.006) and ASIR60 (OR = 1.06; 95% CI = 1.01-1.12; P = 0.034) compared to FBP. Conclusion ASIR significantly decreased the subjectively assessed homogeneity of the left ventricular lumen and increased the objectively measured noise compared to FBP. Considering these results, ASIR at a reduced radiation dose should be implemented with caution.

How Different Iterative and Filtered Back Projection Kernels Affect Computed Tomography Numbers and Low Contrast Detectability.

OBJECTIVE: The aim of this study was to evaluate how different iterative and filtered back projection kernels affect the computed tomography (CT) numbers and low contrast detectability. METHODS: Five different scans were performed at 6 different tube potentials on the same Catphan 600 phantom using approximately the same dose level and otherwise identical settings. The scans were reconstructed using all available filtered back projection body kernels and with iterative reconstruction techniques. RESULTS: The CT numbers and the contrast-to-noise ratios were reported and how they are affected by the kernel choice and strength of iterative reconstruction. CONCLUSIONS: Iterative reconstruction improved contrast-to-noise ratio in most cases, but in certain situations, it decreased it. Variations in CT numbers can be large between kernels with similar sharpness for certain densities.

Image texture and radiation dose properties in CT.

The aim of this study was to compare image noise properties of GE Discovery HD 750 and Toshiba Aquilion ONE. The uniformity section of a Catphan 600 image quality assurance phantom was scanned with both scanners, at different dose levels and with extension rings simulating patients of different sizes. 36 datasets were obtained and analyzed in terms of noise power spectrum. All the results prove that introduction of extension rings significantly altered the image quality with respect to noise properties. Without extension rings, the Toshiba scanner had lower total visible noise than GE (with GE as reference: FC18 had 82% and FC08 had 80% for 10 mGy, FC18 had 77% and FC08 74% for 15 mGy, FC18 had 80% and FC08 77% for 20 mGy). The total visible noise (TVN) for 20 and 15 mGy were similar for the phantom with the smallest additional extension ring, while Toshiba had higher TVN than GE for the 10 mGy dose level (120% FC18, 110% FC08). For the second and third ring, the GE images had lower TVN than Toshiba images for all dose levels (Toshiba TVN is greater than 155% for all cases). The results indi-cate that GE potentially has less image noise than Toshiba for larger patients. The Toshiba FC18 kernel had higher TVN than the Toshiba FC08 kernel with additional beam hardening correction for all dose levels and phantom sizes (120%, 107%, and 106% for FC18 compared to 110%, 98%, and 97%, for FC08, for 10, 15 and 20 mGy doses, respectively).

Influence of adaptive statistical iterative reconstruction algorithm on image quality in coronary computed tomography angiography.

BACKGROUND: Coronary computed tomography angiography (CCTA) requires high spatial and temporal resolution, increased low contrast resolution for the assessment of coronary artery stenosis, plaque detection, and/or non-coronary pathology. Therefore, new reconstruction algorithms, particularly iterative reconstruction (IR) techniques, have been developed in an attempt to improve image quality with no cost in radiation exposure. PURPOSE: To evaluate whether adaptive statistical iterative reconstruction (ASIR) enhances perceived image quality in CCTA compared to filtered back projection (FBP). MATERIAL AND METHODS: Thirty patients underwent CCTA due to suspected coronary artery disease. Images were reconstructed using FBP, 30% ASIR, and 60% ASIR. Ninety image sets were evaluated by five observers using the subjective visual grading analysis (VGA) and assessed by proportional odds modeling. Objective quality assessment (contrast, noise, and the contrast-to-noise ratio [CNR]) was analyzed with linear mixed effects modeling on log-transformed data. The need for ethical approval was waived by the local ethics committee as the study only involved anonymously collected clinical data. RESULTS: VGA showed significant improvements in sharpness by comparing FBP with ASIR, resulting in odds ratios of 1.54 for 30% ASIR and 1.89 for 60% ASIR (P = 0.004). The objective measures showed significant differences between FBP and 60% ASIR (P < 0.0001) for noise, with an estimated ratio of 0.82, and for CNR, with an estimated ratio of 1.26. CONCLUSION: ASIR improved the subjective image quality of parameter sharpness and, objectively, reduced noise and increased CNR.

New developed DR detector performs radiographs of hand, pelvic and premature chest anatomies at a lower radiation dose and/or a higher image quality.

A newly developed Digital Radiography (DR) detector has smaller pixel size and higher fill factor than earlier detector models. These technical advantages should theoretically lead to higher sensitivity and higher spatial resolution, thus making dose reduction possible without scarifying image quality compared to previous DR detector versions. To examine whether the newly developed Canon CXDI-70C DR detector provides an improved image quality and/or allows for dose reductions in hand and pelvic bone examinations as well as premature chest examinations, compared to the previous (CXDI-55C) DR detector version. A total of 450 images of a technical Contrast-Detail phantom were imaged on a DR system employing various kVp and mAs settings, providing an objective image quality assessment. In addition, 450 images of anthropomorphic phantoms were taken and analyzed by three specialized radiologists using Visual Grading Analysis (VGA). The results from the technical phantom studies showed that the image quality expressed as IQFINV values was on average approximately 45 % higher with the CXDI-70C detector compared to the CXDI-55C detector. Consistently, the VGA results from the anatomical phantom studies indicated that by using the CXDI-70C detector, diagnostic image quality could be maintained at a dose reduction of in average 30 %, depending on anatomy and kVp level. This indicates that the CXDI-70C detector is significantly more sensitive than the previous model, and supports a better clinical image quality. By using the newly developed DR detector a significant dose reduction is possible while maintaining image quality.