Artificial Intelligence to Assist in Exclusion of Coronary Atherosclerosis During CCTA Evaluation of Chest Pain in the Emergency Department: Preparing an Application for Real-world Use.

Coronary computed tomography angiography (CCTA) evaluation of chest pain patients in an emergency department (ED) is considered appropriate. While a "negative" CCTA interpretation supports direct patient discharge from an ED, labor-intensive analyses are required, with accuracy in jeopardy from distractions. We describe the development of an artificial intelligence (AI) algorithm and workflow for assisting qualified interpreting physicians in CCTA screening for total absence of coronary atherosclerosis. The two-phase approach consisted of (1) phase 1-development and preliminary testing of an algorithm for vessel-centerline extraction classification in a balanced study population (n = 500 with 50% disease prevalence) derived by retrospective random case selection, and (2) phase 2-simulated clinical Trialing of developed algorithm on a per-case (entire coronary artery tree) basis in a more "real-world" study population (n = 100 with 28% disease prevalence) from an ED chest pain series. This allowed pre-deployment evaluation of the AI-based CCTA screening application which provides vessel-by-vessel graphic display of algorithm inference results integrated into a clinically capable viewer. Algorithm performance evaluation used area under the receiver operating characteristic curve (AUC-ROC); confusion matrices reflected ground truth vs AI determinations. The vessel-based algorithm demonstrated strong performance with AUC-ROC = 0.96. In both phase 1 and phase 2, independent of disease prevalence differences, negative predictive values at the case level were very high at 95%. The rate of completion of the algorithm workflow process (96% with inference results in 55-80 s) in phase 2 depended on adequate image quality. There is potential for this AI application to assist in CCTA interpretation to help extricate atherosclerosis from chest pain presentations.

Are quantitative features of lung nodules reproducible at different CT acquisition and reconstruction parameters?

Consistency and duplicability in Computed Tomography (CT) output is essential to quantitative imaging for lung cancer detection and monitoring. This study of CT-detected lung nodules investigated the reproducibility of volume-, density-, and texture-based features (outcome variables) over routine ranges of radiation dose, reconstruction kernel, and slice thickness. CT raw data of 23 nodules were reconstructed using 320 acquisition/reconstruction conditions (combinations of 4 doses, 10 kernels, and 8 thicknesses). Scans at 12.5%, 25%, and 50% of protocol dose were simulated; reduced-dose and full-dose data were reconstructed using conventional filtered back-projection and iterative-reconstruction kernels at a range of thicknesses (0.6-5.0 mm). Full-dose/B50f kernel reconstructions underwent expert segmentation for reference Region-Of-Interest (ROI) and nodule volume per thickness; each ROI was applied to 40 corresponding images (combinations of 4 doses and 10 kernels). Typical texture analysis metrics (including 5 histogram features, 13 Gray Level Co-occurrence Matrix, 5 Run Length Matrix, 2 Neighboring Gray-Level Dependence Matrix, and 3 Neighborhood Gray-Tone Difference Matrix) were computed per ROI. Reconstruction conditions resulting in no significant change in volume, density, or texture metrics were identified as "compatible pairs" for a given outcome variable. Our results indicate that as thickness increases, volumetric reproducibility decreases, while reproducibility of histogram- and texture-based features across different acquisition and reconstruction parameters improves. To achieve concomitant reproducibility of volumetric and radiomic results across studies, balanced standardization of the imaging acquisition parameters is required.

Performance of a Deep Neural Network Algorithm Based on a Small Medical Image Dataset: Incremental Impact of 3D-to-2D Reformation Combined with Novel Data Augmentation, Photometric Conversion, or Transfer Learning.

Collecting and curating large medical-image datasets for deep neural network (DNN) algorithm development is typically difficult and resource-intensive. While transfer learning (TL) decreases reliance on large data collections, current TL implementations are tailored to two-dimensional (2D) datasets, limiting applicability to volumetric imaging (e.g., computed tomography). Targeting performance enhancement of a DNN algorithm based on a small image dataset, we assessed incremental impact of 3D-to-2D projection methods, one supporting novel data augmentation (DA); photometric grayscale-to-color conversion (GCC); and/or TL on training of an algorithm from a small coronary computed tomography angiography (CCTA) dataset (200 examinations, 50% with atherosclerosis and 50% atherosclerosis-free) producing 245 diseased and 1127 normal coronary arteries/branches. Volumetric CCTA data was converted to a 2D format creating both an Aggregate Projection View (APV) and a Mosaic Projection View (MPV), supporting DA per vessel; both grayscale and color-mapped versions of each view were also obtained. Training was performed both without and with TL, and algorithm performance of all permutations was compared using area under the receiver operating characteristics curve. Without TL, APV performance was 0.74 and 0.87 on grayscale and color images, respectively, compared to 0.90 and 0.87 for MPV. With TL, APV performance was 0.78 and 0.88 on grayscale and color images, respectively, compared with 0.93 and 0.91 for MPV. In conclusion, TL enhances performance of a DNN algorithm from a small volumetric dataset after proposed 3D-to-2D reformatting, but additive gain is achieved with application of either GCC to APV or the proposed novel MPV technique for DA.

A User Interface for Optimizing Radiologist Engagement in Image Data Curation for Artificial Intelligence.

PURPOSE: To delineate image data curation needs and describe a locally designed graphical user interface (GUI) to aid radiologists in image annotation for artificial intelligence (AI) applications in medical imaging. MATERIALS AND METHODS: GUI components support image analysis toolboxes, picture archiving and communication system integration, third-party applications, processing of scripting languages, and integration of deep learning libraries. For clinical AI applications, GUI components included two-dimensional segmentation and classification; three-dimensional segmentation and quantification; and three-dimensional segmentation, quantification, and classification. To assess radiologist engagement and performance efficiency associated with GUI-related capabilities, image annotation rate (studies per day) and speed (minutes per case) were evaluated in two clinical scenarios of varying complexity: hip fracture detection and coronary atherosclerotic plaque demarcation and stenosis grading. RESULTS: For hip fracture, 1050 radiographs were annotated over 7 days (150 studies per day; median speed: 10 seconds per study [interquartile range, 3-21 seconds per study]). A total of 294 coronary CT angiographic studies with 1843 arteries and branches were annotated for atherosclerotic plaque over 23 days (15.2 studies [80.1 vessels] per day; median speed: 6.08 minutes per study [interquartile range, 2.8-10.6 minutes per study] and 73 seconds per vessel [interquartile range, 20.9-155 seconds per vessel]). CONCLUSION: GUI-component compatibility with common image analysis tools facilitates radiologist engagement in image data curation, including image annotation, supporting AI application development and evolution for medical imaging. When complemented by other GUI elements, a continuous integrated workflow supporting formation of an agile deep neural network life cycle results.Supplemental material is available for this article.© RSNA, 2019.

Aortic volume as an indicator of disease progression in patients with untreated infrarenal abdominal aneurysm.

OBJECTIVE: The maximal diameter of an abdominal aortic aneurysm (AAA) and the change in diameter over time reflect rupture risk and are used for surgical planning. However, evidence has emerged that aneurysm volume may be a better indicator of AAA remodeling. The purpose of this study was to assess the relationship between the volume and maximal diameter of the abdominal aorta in patients with untreated infrarenal AAA. MATERIALS AND METHODS: This was a retrospective study of 100 patients with infrarenal AAA who were followed for more than 6 months. We examined 2 sets of computed tomography images for each patient, acquired ≥ 6 months apart. The maximal diameter and volume of the infrarenal abdominal aorta were determined by semiautomated segmentation software. RESULTS: At baseline, mean maximal infrarenal diameter was 5.1 ± 1.0 cm and mean aortic volume was 139 ± 72 mL. There was good correlation between the maximal diameter and aortic volume at baseline (r(2) = 0.55; P<0.001). The mean change in maximal diameter between studies was 0.2 ± 0.3 cm and the mean volume change was 19 ± 19 mL. However, the correlation between diameter change and volume change was modest (r(2) = 0.34; P=0.001). Most patients (n = 64) had no measurable change in maximal diameter between studies (≤ 2 mm), but the change in volume was found to vary widely (-2 to 69 mL). CONCLUSION: In patients with untreated infrarenal AAA, a change in aortic volume can occur in the absence of a significant change in maximal diameter. Additional work is needed to examine the relationship between change in AAA volume and outcomes in this patient group.

Influence of coronary artery stenosis severity and coronary collateralization on extent of chronic myocardial scar: insights from quantitative coronary angiography and delayed-enhancement MRI.

OBJECTIVES: In patients with chronic ischemic heart disease, the relationship between coronary artery lesion severity and myocardial scarring is unknown.The purpose of this study was to examine the relationship between proximal coronary artery stenosis severity, the amount of coronary collateralization, and myocardial scar extent in the distal distribution of the affected coronary artery based on both quantitative coronary angiography (QCA) and delayed-enhancement magnetic resonance imaging (DE-MRI). METHODS: Thirty-four patients (26 males, 8 females; age range: 35-86 years) with a coronary artery containing a single, proximal stenosis >/=30% by quantitative coronary angiography (QCA) underwent DE-MRI. The relationship between stenosis severity, collateralization, and myocardial scar morphology (area, transmurality and patchiness) was examined using linear mixed-model ANCOVA. RESULTS: There was a statistically significant correlation between stenosis severity and scar extent (r=0.53, p<0.01). Patients with hemodynamically significant stenoses (>/=70%) exhibited significantly greater collateralization (p<0.05) and scar extent (p<0.01) than patients with <70% stenosis. However, scarring was often found in patients with stenoses <70%. Also, greater stenosis severity (93+/-14%) and mean scar extent (41+/-35%) were found in patients with collaterals than in patients without collaterals (diameter stenosis 48+/-10%, p<0.01) (scar extent 19+/-29%, p=0.01). CONCLUSIONS: Using QCA and DE-MRI, we demonstrate a significant relationship between coronary artery stenosis severity and myocardial scar extent, in the absence of a documented history of acute infarction. The relationship likely reflects increasing ischemia leading to scar formation in the range of angiographically significant stenosis. However, in the absence of collateralization, scar was observed without significant stenosis, especially in females.

Prescription or proscription? The general failure of attempts to litigate and legislate against PBMS as "fiduciaries," and the role of market forces allowing PBMS to contain private-sector prescription drug prices.

Pharmacy benefit managers (PBMs), which generally administer prescription drug benefits as one component of an employer's or other sponsor's health insurance plan, have come under fire in recent years for turning profits at a time when consumer advocates and employers are struggling to contain the costs of health insurance and prescription drugs. Lawsuits alleging that PBMs are breaching certain fiduciary duties to the health plans they serve, however, have failed for the most part on grounds that PBMs are not "fiduciaries" under the Employee Retirement Income Security Act (ERISA). Moreover, states' attempts to regulate PBMs through legislation imposing fiduciary obligations and other related requirements have also generally failed for many different reasons. This Article examines the PBM industry, recent legal developments concerning PBMs' status as ERISA "fiduciaries", the arguments being made for and against stricter regulation of PBMs' business practices, and why litigation and legislation attempting to impose fiduciary obligations upon PBMs have generally failed. The authors conclude that it is market forces and competition, rather than litigation or legislation, that will effectively motivate PBMs to play a role in the cost containment of prescription drugs in the years ahead.

Prevalence and correlates of septal delayed contrast enhancement in patients with pulmonary hypertension.

Using cardiac magnetic resonance, the presence of myocardial delayed contrast enhancement (DCE) has been described in the ventricular septum at the level of the right ventricular insertion points in patients with pulmonary hypertension (PH). The aim of this study was to investigate the prevalence, extent, and correlates of this finding. Septal DCE was evaluated in 55 patients with known or suspected PH of various causes. The extent of DCE was estimated visually with an insertion enhancement score (range 0 to 4) and quantified as DCE mass. The results were correlated with cine magnetic resonance and right-sided cardiac catheterization. Predictors of DCE were investigated using multivariate analysis. PH at rest was present in 42 patients (group 1) and absent in 13 (group 2). DCE was noted in 41 patients (97%) in group 1 and 3 (23%) in group 2 (p <0.0001). The extent of DCE was higher in group 1 than group 2 (median insertion enhancement score 3 vs 0, median DCE mass 8.7 vs 0 g, respectively; p <0.0001 for both). The extent of DCE showed moderate to good univariate correlations (r = 0.5 to 0.73) with pulmonary pressures and with right ventricular volumes, mass, and ejection fractions. In multivariate analysis, systolic pulmonary pressure was the only predictor of DCE. In conclusion, the presence of septal DCE at the right ventricular insertion points is common in PH of different causes, and the level of systolic pulmonary pressure elevation appears to be the main determinant of this finding.

Coregistered MR imaging myocardial viability maps and multi-detector row CT coronary angiography displays for surgical revascularization planning: initial experience.

PURPOSE: To evaluate assignment of left ventricular (LV) myocardial segments to coronary arterial territories by using coregistered magnetic resonance (MR) imaging and multi-detector row computed tomography (CT) displays; to assess the accuracy of coregistered displays in determining the distribution of clinically important coronary artery disease (CAD) and regional effect of CAD on LV myocardium in patients with chronic ischemic heart disease (CIHD); and to determine the utility of coregistered displays in optimizing surgical revascularization planning. MATERIALS AND METHODS: This study was HIPAA compliant and was approved by the local Institutional Review Board, with waiver of informed consent. Twenty-six patients (19 men, seven women; age, 56 years +/- 12 [+/- standard deviation]) with CIHD underwent MR imaging assessment of myocardial viability and multi-detector row CT assessment of CAD on the same day. For coregistration, a population-based LV model was fit to each data set separately; models were then registered spatially. For data analysis, correspondence between coregistered displays and the 17-segment LV model for assessment of CIHD was evaluated, accuracy of using coregistered displays to evaluate the extent of CAD and myocardial disease was assessed, and utility of coregistered displays in optimizing surgical revascularization planning was determined. RESULTS: Coronary assignment for coregistered displays and the 17-segment LV model differed in 17% of myocardial segments. For the majority of patients, three segments (midanterolateral [62%], apical lateral [73%], and apical inferior [58%]) were discordant. Segments were supplied by the left anterior descending artery, a diagonal branch, or a ramus intermedius with diagonal distribution in all but one case. Coregistered displays were deemed concordant with selective coronary angiography and alternate myocardial imaging in all cases. Overall, surgical planning was potentially enhanced in 83% of cases because, compared with alternate imaging modalities, coregistered displays were believed to demonstrate the relationship between coronary arteries and underlying myocardial tissue more definitively and efficiently (for patients in whom surgery was performed) or more correctly and comprehensively (for a presumably better-tailored surgery). CONCLUSION: Assessment of CIHD can be improved by using coregistered displays that directly relate the condition of LV myocardium to the anatomy of the coronary arteries in individual patients.

Segmentation of non-viable myocardium in delayed enhancement magnetic resonance images.

PURPOSE: To evaluate six algorithms for segmenting non-viable left ventricular (LV) myocardium in delayed enhancement (DE) magnetic resonance imaging (MRI). METHODS: Twenty-three patients with known chronic ischemic heart disease underwent DE-MRI. DE images were first manually thresholded using an interactive region-filling tool to isolate non-viable myocardium. Then, six thresholding algorithms, based on the image intensity characteristics of either LV blood pool (BP), viable LV myocardium, or both, were applied to each image. For the Mean-2SD(BP) algorithm, thresholds were equal to the mean BP intensity minus twice its standard deviation. For the Mean + 2SD(Semi), Mean + 3SD(Semi), Mean + 2SD(Auto), and Mean + 3SD(Auto) algorithms, thresholds equaled the mean intensity of viable myocardium plus twice (or thrice, as denoted by the name) the standard deviation of intensity (subscripts denote how these values were determined: automatic or semi-automatic). For the Minimum Intensity algorithm, the threshold equaled the minimum intensity between the BP and LV myocardium mean intensities. Percent Scar was defined as the ratio of non-viable to total myocardial pixels in each image. Agreement between each algorithm and manual thresholding was assessed using Bland-Altman analysis. RESULTS: Mean Percent Scar was 25 +/- 16% by manual thresholding. Five of the six algorithms demonstrated mean bias within +/-3% (all except Mean+2SD(Auto)); however, limits of agreement (LoA) were large in general (range 12-36%). The best overall agreement was demonstrated by the Mean + 2SD(Semi) (bias, 0%; LoA, 12%) and Mean + 3SD(Semi)(bias, -3%; LoA, 14%) algorithms. CONCLUSION: On average, five of the six algorithms proved satisfactory for clinical implementation; however, in some images, manual correction of automatic results was necessary.

Quantitative assessment of myocardial scar in delayed enhancement magnetic resonance imaging.

PURPOSE: To characterize the extent and distribution of left ventricular myocardial scar in delayed enhancement magnetic resonance imaging (MRI). MATERIALS AND METHODS: Delayed enhancement images from 18 patients were categorized into three groups based on myocardial scar appearance: discrete myocardial infarction (N = 10), diffuse fibrosis (N = 4), and circumferential endocardial scarring (N = 4). Images were segmented manually by two observers (twice by one observer) to identify nonviable myocardium. Scar was characterized by the following morphologic parameters: the relative area of nonviable myocardium (Percent Scar); a measure of scar cohesion (Patchiness); and the extent to which scar traversed the ventricle wall (Trans>50). RESULTS: The three scar parameters successfully discriminated between patient groups, although no one parameter was able to differentiate between all groups. The average bias between readers was approximately 3% for each parameter, and the average bias between repeated measurements was 1%. In addition, five patients exhibited regions of nonhyperenhanced nonviable myocardium that were expected to show hyperenhancement based upon their location within the infarct zone and appearance on cine images. CONCLUSION: Quantitative characterization of myocardial scar showed good interobserver and intraobserver agreement. However, the appearance of nonhyperenhanced scar in chronic ischemia is problematic for segmentation of delayed enhancement images.