Anatomy-Guided Weakly-Supervised Abnormality Localization in Chest X-rays.

Creating a large-scale dataset of abnormality annotation on medical images is a labor-intensive and costly task. Leveraging weak supervision from readily available data such as radiology reports can compensate lack of large-scale data for anomaly detection methods. However, most of the current methods only use image-level pathological observations, failing to utilize the relevant anatomy mentions in reports. Furthermore, Natural Language Processing (NLP)-mined weak labels are noisy due to label sparsity and linguistic ambiguity. We propose an Anatomy-Guided chest X-ray Network (AGXNet) to address these issues of weak annotation. Our framework consists of a cascade of two networks, one responsible for identifying anatomical abnormalities and the second responsible for pathological observations. The critical component in our framework is an anatomy-guided attention module that aids the downstream observation network in focusing on the relevant anatomical regions generated by the anatomy network. We use Positive Unlabeled (PU) learning to account for the fact that lack of mention does not necessarily mean a negative label. Our quantitative and qualitative results on the MIMIC-CXR dataset demonstrate the effectiveness of AGXNet in disease and anatomical abnormality localization. Experiments on the NIH Chest X-ray dataset show that the learned feature representations are transferable and can achieve the state-of-the-art performances in disease classification and competitive disease localization results. Our code is available at https://github.com/batmanlab/AGXNet.

Retraining an open-source pneumothorax detecting machine learning algorithm for improved performance to medical images.

PURPOSE: To validate a machine learning model trained on an open source dataset and subsequently optimize it to chest X-rays with large pneumothoraces from our institution. METHODS: The study was retrospective in nature. The open-source chest X-ray (CXR8) dataset was dichotomized to cases with pneumothorax (PTX) and all other cases (non-PTX), resulting in 41,946 non-PTX and 4696 PTX cases for the training set and 11,120 non-PTX and 541 PTX cases for the validation set. A limited supervision machine learning model was constructed to incorporate both localized and unlocalized pathology. Cases were then queried from our health system from 2013 to 2017. A total of 159 pneumothorax and 682 non-pneumothorax cases were available for the training set. For the validation set, 48 pneumothorax and 1287 non-pneumothorax cases were available. The model was trained, a receiver operator curve (ROC) was created, and output metrics, including area under the curve (AUC), sensitivity and specificity were calculated. RESULTS: Initial training of the model using the CXR8 dataset resulted in an AUC of 0.90 for pneumothorax detection. Naively inferring our own validation dataset on the CXR8 trained model output an AUC of 0.59. After re-training the model with our own training dataset, the validation dataset inference output an AUC of 0.90. CONCLUSION: Our study showed that even though you may get great results on open-source datasets, those models may not translate well to real world data without an intervening retraining process.

Anomalous Left-sided Superior Vena Cava with Cephalad Flow.

The case of a 24-year-old male with complaints of migraine headaches was referred for echocardiography. The rest of medical history was unremarkable. Agitated saline contrast bubble study showed evidence of a right to left intracardiac shunt, probably secondary to a patent foramen ovale. Results of a transesophageal echocardiogram suggested the possibility of an anomalous venous circulation and eventually identified as anomalous left-sided superior vena cava with cardiac magnetic resonance imaging.

Value-Based Assessment of Radiology Reporting Using Radiologist-Referring Physician Two-Way Feedback System-a Design Thinking-Based Approach.

In the era of value-based healthcare, many aspects of medical care are being measured and assessed to improve quality and reduce costs. Radiology adds enormously to health care costs and is under pressure to adopt a more efficient system that incorporates essential metrics to assess its value and impact on outcomes. Most current systems tie radiologists' incentives and evaluations to RVU-based productivity metrics and peer-review-based quality metrics. In a new potential model, a radiologist's performance will have to increasingly depend on a number of parameters that define "value," beginning with peer review metrics that include referrer satisfaction and feedback from radiologists to the referring physician that evaluates the potency and validity of clinical information provided for a given study. These new dimensions of value measurement will directly impact the cascade of further medical management. We share our continued experience with this project that had two components: RESP (Referrer Evaluation System Pilot) and FRACI (Feedback from Radiologist Addressing Confounding Issues), which were introduced to the clinical radiology workflow in order to capture referrer-based and radiologist-based feedback on radiology reporting. We also share our insight into the principles of design thinking as applied in its planning and execution.

Technical Challenges in the Clinical Application of Radiomics.

Radiomics is a quantitative approach to medical image analysis targeted at deciphering the morphologic and functional features of a lesion. Radiomic methods can be applied across various malignant conditions to identify tumor phenotype characteristics in the images that correlate with their likelihood of survival, as well as their association with the underlying biology. Identifying this set of characteristic features, called tumor signature, holds tremendous value in predicting the behavior and progression of cancer, which in turn has the potential to predict its response to various therapeutic options. We discuss the technical challenges encountered in the application of radiomics, in terms of methodology, workflow integration, and user experience, that need to be addressed to harness its true potential.

RV-pulmonary arterial coupling predicts outcome in patients referred for pulmonary hypertension.

OBJECTIVE: Prognosis in pulmonary hypertension (PH) is largely determined by RV function. However, uncertainty remains about what metrics of RV function might be most clinically relevant. The purpose of this study was to assess the clinical relevance of metrics of RV functional adaptation to increased afterload. METHODS: Patients referred for PH underwent right heart catheterisation and RV volumetric assessment within 48 h. A RV maximum pressure (Pmax) was calculated from the RV pressure curve. The adequacy of RV systolic functional adaptation to increased afterload was estimated either by a stroke volume (SV)/end-systolic volume (ESV) ratio, a Pmax/mean pulmonary artery pressure (mPAP) ratio, or by EF (RVEF). Diastolic function of the RV was estimated by a diastolic elastance coefficient ß. Survival analysis was via Cox proportional HR, and Kaplan-Meier with the primary outcome of time to death or lung transplant. RESULTS: Patients (n=50; age 58±13 yrs) covered a range of mPAP (13-79 mm Hg) with an average RVEF of 39±17% and ESV of 143±89 mL. Average estimates of the ratio of end-systolic ventricular to arterial elastance were 0.79±0.67 (SV/ESV) and 2.3±0.65 (Pmax/mPAP-1). Transplantation-free survival was predicted by right atrial pressure, mPAP, pulmonary vascular resistance, ß, SV, ESV, SV/ESV and RVEF, but after controlling for right atrial pressure, mPAP, and SV, SV/ESV was the only independent predictor. CONCLUSIONS: The adequacy of RV functional adaptation to afterload predicts survival in patients referred for PH. Whether this can simply be evaluated using RV volumetric imaging will require additional confirmation.

Head CT scan overuse in frequently admitted medical patients.

BACKGROUND: Patients frequently admitted to medical services undergo extensive computed tomography (CT) imaging. Some of this imaging may be unnecessary, and in particular, head CT scans may be over-used in this patient population. We describe the frequency of abnormal head CT scans in patients with multiple medical hospitalizations. METHODS: We retrospectively reviewed all CT scans done in 130 patients with 7 or more admissions to medical services between January 1 and December 31, 2011 within an integrated health care system. We calculated the number of CT scans, anatomic site of imaging, and source of ordering (emergency department, inpatient floor). We scored all head CT scans on a 0-4 scale based on the severity of radiographic findings. Higher scores signified more clinically important findings. RESULTS: There were 795 CT scans performed in total, with a mean of 6.7 (± SD 5.8) CT scans per patient. Abdominal/pelvis (39%), chest (30%), and head (22%) CT scans were the most frequently obtained. The mean number of head CT scans performed was 2.9 (SD ± 4.2). Inpatient floors were the major site of CT scan ordering (53.7%). Of 172 head CT scans, only 4% had clinically significant findings (scores of 3 or 4). CONCLUSIONS: Patients with frequent medical admissions are medically complex and undergo multiple CT scans in a year. The vast majority of head CT scans lack clinically significant findings and should be ordered less frequently. Interdisciplinary measures should be advocated by hospitalists, emergency departments, and radiologists to decrease unnecessary imaging in this population.

Clinical social networking--a new revolution in provider communication and delivery of clinical information across providers of care?

The adoption of social media technologies appears to enhance clinical outcomes through improved communications as reported by Bacigalupe (Fam Syst Heal 29(1):1-14, 2011). The ability of providers to more effectively, directly, and rapidly communicate among themselves as well as with patients should strengthen collaboration and treatment as reported by Bacigalupe (Fam Syst Heal 29(1):1-14, 2011). This paper is a case study in one organization's development of an internally designed and developed social technology solution termed "Unite." The Unite system combines social technologies' features including push notifications, messaging, community groups, and user lists with clinical workflow and applications to construct dynamic provider networks, simplify communications, and facilitate clinical workflow optimization. Modeling Unite as a social technology may ease adoption barriers. Developing a social network that is integrated with healthcare information systems in the clinical space opens the doors to capturing and studying the way in which providers communicate. The Unite system appears to have the potential to breaking down existing communication paradigms. With Unite, a rich set of usage data tied to clinical events may unravel alternative networks that can be leveraged to advance patient care.

Free-breathing, motion-corrected late gadolinium enhancement is robust and extends risk stratification to vulnerable patients.

BACKGROUND: Routine clinical use of novel free-breathing, motion-corrected, averaged late-gadolinium-enhancement (moco-LGE) cardiovascular MR may have advantages over conventional breath-held LGE (bh-LGE), especially in vulnerable patients. METHODS AND RESULTS: In 390 consecutive patients, we collected bh-LGE and moco-LGE with identical image matrix parameters. In 41 patients, bh-LGE was abandoned because of image quality issues, including 10 with myocardial infarction. When both were acquired, myocardial infarction detection was similar (McNemar test, P=0.4) with high agreement (κ=0.95). With artifact-free bh-LGE images, pixelwise myocardial infarction measures correlated highly (R(2)=0.96) without bias. Moco-LGE was faster, and image quality and diagnostic confidence were higher on blinded review (P<0.001 for all). During a median of 1.2 years, 20 heart failure hospitalizations and 18 deaths occurred. For bh-LGE, but not moco-LGE, inferior image quality and bh-LGE nonacquisition were linked to patient vulnerability confirmed by adverse outcomes (log-rank P<0.001). Moco-LGE significantly stratified risk in the full cohort (log-rank P<0.001), but bh-LGE did not (log-rank P=0.056) because a significant number of vulnerable patients did not receive bh-LGE (because of arrhythmia or inability to hold breath). CONCLUSIONS: Myocardial infarction detection and quantification are similar between moco-LGE and bh-LGE when bh-LGE can be acquired well, but bh-LGE quality deteriorates with patient vulnerability. Acquisition time, image quality, diagnostic confidence, and the number of successfully scanned patients are superior with moco-LGE, which extends LGE-based risk stratification to include patients with vulnerability confirmed by outcomes. Moco-LGE may be suitable for routine clinical use.

Use of cardiac CT angiography imaging in an epidemiology study - the Methodology of the Multicenter AIDS Cohort Study cardiovascular disease substudy.

OBJECTIVE: The methodology for use of cardiac CT angiography (CTA) in low risk populations is not well defined. In order to present a reference for future studies, we present CTA methodology that is being used in an epidemiology study- the Multicenter AIDS Cohort Study (MACS). METHODS: The Multicenter AIDS Cohort Study (MACS) is an on-going multicenter prospective, observational cohort study. The MACS Cardiovascular Disease substudy plans to enroll 800 men (n= 575 HIV seropositive and n=225 HIV seronegative) age 40-70 years for coronary atherosclerosis imaging using cardiac CTA. The protocol includes heart rate (HR) optimization with beta- blockers; use of proper field of view; scan length limitation; prospective ECG-gating using the lowest beam voltage possible. All scans are evaluated for presence, extent, and composition of coronary atherosclerosis, left atrial volumes, left ventricular volume and mass and non-coronary cardiac pathology. RESULTS: The first 498 participants had an average radiation dose of 2.5±1.6 milliSieverts (mSv) for the cardiac CTA study. Overall quality of scans was fair to excellent in 98.6% of studies. There were three significant adverse events-two allergic reactions to contrast and one subcutaneous contrast extravasation. CONCLUSION: Cardiac CTA was safe and afforded a low effective radiation exposure to these asymptomatic research participants and provides valuable cardiovascular endpoints for scientific analysis. The cardiac CTA methodology described here may serve as a reference for use in future epidemiology studies aiming to assess coronary atherosclerosis and cardiac anatomy in low risk populations while minimizing radiation exposure.

Workflow continuity--moving beyond business continuity in a multisite 24-7 healthcare organization.

As hospitals move towards providing in-house 24 × 7 services, there is an increasing need for information systems to be available around the clock. This study investigates one organization's need for a workflow continuity solution that provides around the clock availability for information systems that do not provide highly available services. The organization investigated is a large multifacility healthcare organization that consists of 20 hospitals and more than 30 imaging centers. A case analysis approach was used to investigate the organization's efforts. The results show an overall reduction in downtimes where radiologists could not continue their normal workflow on the integrated Picture Archiving and Communications System (PACS) solution by 94 % from 2008 to 2011. The impact of unplanned downtimes was reduced by 72 % while the impact of planned downtimes was reduced by 99.66 % over the same period. Additionally more than 98 h of radiologist impact due to a PACS upgrade in 2008 was entirely eliminated in 2011 utilizing the system created by the workflow continuity approach. Workflow continuity differs from high availability and business continuity in its design process and available services. Workflow continuity only ensures that critical workflows are available when the production system is unavailable due to scheduled or unscheduled downtimes. Workflow continuity works in conjunction with business continuity and highly available system designs. The results of this investigation revealed that this approach can add significant value to organizations because impact on users is minimized if not eliminated entirely.

Prevalence of aortic root dilation in patients with CT angiography of the aorta.

PURPOSE: To investigate the prevalence of aortic root dilation in patients who underwent CT angiography of the thoracic aorta. MATERIALS AND METHODS: In 95 patients, 64-slice multislice computed tomography was performed for evaluation of the thoracic aorta. Measurements of the annulus, sinuses of valsalva (SOV), sinotubular junction (STJ), and maximum ascending aorta (AAo) were made by double oblique multiplanar reformation (MPR). For the AAo, STJ, and SOV, dilation was defined as greater than 40 mm; for annulus, the dilation criterion was greater than 27 mm. RESULTS: Overall, 52 patients were diagnosed with a dilated AAo. Of those patients with dilated AAo, 28 patients had a dilated annulus, 27 patients had dilated SOV, and 11 patients had STJ dilation. Forty-three patients presented with normal AAo; 12 patients had annulus dilation; 12 patients had SOV dilation; and 4 patients had STJ dilation. In patients with dilated AAo, 38% also had a dilated annulus, 52% showed SOV dilation, and 21% presented with STJ dilation, compared to 28% annulus dilation, 28% SOV dilation, and 9% STJ dilation in patients with an AAo of normal caliber. CONCLUSION: Our data indicate a higher prevalence of aortic root dilation among patients with dilated AAo.

The aortic root: comparison of measurements from ECG-gated CT angiography with transthoracic echocardiography.

PURPOSE: The purpose of this study was to compare the measurements of the aortic root obtained from electrocardiographically (ECG)-gated computed tomography (CT) angiography (CTA) to the measurements obtained from transthoracic echocardiography (TTE). MATERIALS AND METHODS: This was a retrospective study in a patient population scanned at our institution between December 2005 and January 2007 with retrospectively ECG-gated CTA. ECG-gated CTA was performed with a 64-section helical CT scanner (Light speed, VCT, GE, Milwaukee, WI). Sixty-eight patients; 51 men and 17 women were included in this study. Aortic root diameters were measured by using double oblique reconstruction from axial source images. The TTE measurements of the aortic root were obtained from the reports that were performed within 2 months of CTA. RESULTS: The average aortic root diameter measured by TTE was 33+/-4.1 mm; on CTA it was 36.9+/-3.8 mm. The median difference between the 2 measurements was 3.9 mm which was significant (P<0.0001). In patients whose aortic root measurements with CTA were normal, the TTE measurements were also normal. However, in the group of patients with dilated aortic roots by CTA, TTE measurements were significantly lower and many were normal. In the group of patients with dilated aortic root by TTE, the CTA measurements of the aortic root were similarly increased. CONCLUSIONS: Retrospective comparison of TTE and CTA measurements of the aortic root reveal that TTE measurements are substantially lower or even normal in patients found to have dilated aortic root by CTA.

Phenotyping the right ventricle in patients with pulmonary hypertension.

Right ventricular (RV) failure is associated with poor outcomes in pulmonary hypertension (PH). We sought to phenotype the RV in PH patients with compensated and decompensated RV function by quantifying regional and global RV structural and functional changes. Twenty-two patients (age 51 +/- 11, 14 females, mean pulmonary artery (PA) pressure range 13-79 mmHg) underwent right heart catheterization, echocardiography, and ECG-gated multislice computed tomography of the chest. Patients were divided into three groups: Normal, PH with hemodynamically compensated, and decompensated RV function (PH-C and PH-D, respectively). RV wall thickness (WT) was measured at end-diastole (ED) and end-systole (ES) in three regions: infundibulum, lateral free wall, and inferior free wall. Globally, RV volumes progressively increased from Normal to PH-C to PH-D and RV ejection fraction decreased. Regionally, WT increased and fractional wall thickening (FWT) decreased in a spatially heterogeneous manner. Infundibular wall stress was elevated and FWT was lower regardless of the status of global RV function. In PH, there are significant phenotypic abnormalities in the RV even in the absence of overt hemodynamic RV decompensation. Regional changes in RV structure and function may be early markers of patients at risk for developing RV failure.

Extra Atrial Disease in Patients with "Lone" Atrial Fibrillation.

AIMS: Lone atrial fibrillation (LAF) is considered by some to be a primary atrial electrophysiologic disorder. However, we have frequently observed evidence of "extraatrial" diseases - atherosclerosis and associated metabolic disorders - in our LAF patients. We sought to characterize and quantify extraatrial disease burden in LAF patients, and to correlate this burden with features of the arrhythmia including pattern (paroxysmal versus persistent) and response to catheter ablation. METHODS AND RESULTS: Forty-six consecutive patients with non-familial LAF underwent assessment for evidence of atherosclerosis (computed tomographic vascular calcification and elevated arterial pulse wave velocity) and associated metabolic diseases (dyslipidemia, insulin resistance and inflammation), and then catheter ablation. The cohort had a significant incidence of atherosclerosis (57%) and metabolic (70%) diseases. Patients with persistent AF tended to have a greater extraatrial disease burden than those with paroxysmal AF. A significant inverse relationship between the rate of ablation success and extraatrial disease burden was demonstrated. CONCLUSIONS: Extraatrial disease was common in this LAF cohort. Correlations between extraatrial disease burden and features of the arrhythmia would, if verified, challenge the notion that LAF is a "primary" electrophysiologic disorder.

Dynamic multidimensional imaging of the human left atrial appendage.

AIMS: The human left atrial appendage (LAA) is a region of increasing interest as a target for intervention. We sought to improve insight into the anatomy of this region using computed tomography (CT). METHODS AND RESULTS: Multidimensional cardiac reconstruction (whole heart and isolated left atrium) from CT images was performed in each of three groups: (i) patients without atrial fibrillation (AF, n =10); (ii) patients with intermittent (paroxysmal) AF (n = 25); (iii) patients with continuous (persistent) AF (n = 10). Indices included LAA morphology, anatomical relationships, dimensions, angulation, and motility. There was substantial interindividual variation in each index. LAA morphologic differences were associated with variations in anatomical relationships. LAA dimensions in AF patients exceeded those in patients without AF, but angulation and motility were similar. The LAA could be subdivided into proximal and distal portions, each of which had distinct morphology and anatomical relationships. Dimensions in men tended to exceed those in women. CONCLUSION: Regardless of AF history, there is broad variation in LAA morphology, anatomical relationships, dimensions, angulation, and motility. These observations may have importance for the development of technologies for therapy delivery in this region.

CT of the pulmonary veins.

Atrial fibrillation (AF) is a common cardiac rhythm disturbance and its incidence is increasing. Radiofrequency catheter ablation (RFCA) is a highly successful therapy for treating AF, and its use is becoming more widespread; however, with its increasing use and evolving technique, known complications are better understood and new complications are emerging. Computed tomography (CT) of the pulmonary veins, or more correctly, the posterior left atrium (LA), has an established role in precisely defining the complex anatomy of the LA and pulmonary veins preablation and has an expanding role in identifying the myriad of possible complications postablation. The purposes of this article are: to review AF and RFCA; to discuss CT evaluation of the LA and pulmonary veins preablation; and to review the complications of RFCA focusing on the role of CT postablation.

Foot and ankle biomechanics.

With the advent of imaging advancements there has been renewed interest in the foot and ankle. However, many of the basic functions and biomechanical considerations of the bones, joints, and specialized tissues of the normal and dysfunctional foot and ankle remain unfamiliar to many radiologists. This article focuses on the basic biomechanics, normal alignment, and common alignment disorders of the foot and ankle that are relevant to radiologists.

Polyethylene glycol diisocyanate decreases platelet deposition after balloon injury of rabbit femoral arteries.

BACKGROUND: Platelet deposition after angioplasty remains problematic and may contribute to intimal hyperplasia and restenosis. We proposed that polyethylene glycol diisocyanate (PEG-DISO), a polymer that rapidly forms covalent linkages with amine residues on proteins, could mask thrombogenic vascular wall proteins from platelets, thereby abrogating acute platelet deposition. METHODS AND RESULTS: To test this hypothesis, we isolated the femoral arteries of 10 New Zealand White rabbits and injured them with 3 passes of a 2F Fogarty catheter which was inserted through a distal arteriotomy. Immediately after balloon injury, (111)indium-labeled autologous platelets were infused peripherally and the injured femoral arteries were randomly treated for 1 minute with a PEG-DISO solution in one artery and a control solution of the phosphate buffered saline vehicle in the contralateral artery. Following treatment, reflow was initiated. The vessels were harvested after 1 hour and radioactivity was quantified in a gamma counter. Platelet counts were standardized by weight and expressed as platelets/mg (mean +/- SEM). Platelet deposition onto arteries treated with PEG-DISO was (1.2 +/- 0.5) x 10(6) platelets/mg compared to (5.6 +/- 4.2) x 10(6) platelets/mg onto the contralateral control arteries treated with vehicle (P < 0.005). Scanning electron micrographs of the injured vessel segment confirmed qualitatively less platelet deposition on the treated segments than on the control segments. CONCLUSION: Treatment with PEG-DISO significantly inhibited platelet deposition after vascular injury. These data support the hypothesis that treatment with PEG-DISO masks surface adhesive proteins from platelet receptors in vivo and that the resulting molecular barrier significantly reduces platelet deposition onto the damaged vessel wall for at least one hour. The formation of a molecularly thin barrier to platelet deposition may thus be a novel and effective treatment to abrogate acute intravascular thrombosis and may have value in the treatment of restenosis.