Deep Learning for Detection of Pneumothorax and Pleural Effusion on Chest Radiographs: Validation Against Computed Tomography, Impact on Resident Reading Time, and Interreader Concordance.

PURPOSE: To study the performance of artificial intelligence (AI) for detecting pleural pathology on chest radiographs (CXRs) using computed tomography as ground truth. PATIENTS AND METHODS: Retrospective study of subjects undergoing CXR in various clinical settings. Computed tomography obtained within 24 hours of the CXR was used to volumetrically quantify pleural effusions (PEfs) and pneumothoraxes (Ptxs). CXR was evaluated by AI software (INSIGHT CXR; Lunit) and by 3 second-year radiology residents, followed by AI-assisted reassessment after a 3-month washout period. We used the area under the receiver operating characteristics curve (AUROC) to assess AI versus residents' performance and mixed-model analyses to investigate differences in reading time and interreader concordance. RESULTS: There were 96 control subjects, 165 with PEf, and 101 with Ptx. AI-AUROC was noninferior to aggregate resident-AUROC for PEf (0.82 vs 0.86, P < 0.001) and Ptx (0.80 vs 0.84, P = 0.001) detection. AI-assisted resident-AUROC was higher but not significantly different from the baseline. AI-assisted reading time was reduced by 49% (157 vs 80 s per case, P = 0.009), and Fleiss kappa for Ptx detection increased from 0.70 to 0.78 ( P = 0.003). AI decreased detection error for PEf (odds ratio = 0.74, P = 0.024) and Ptx (odds ratio = 0.39, P < 0.001). CONCLUSION: Current AI technology for the detection of PEf and Ptx on CXR was noninferior to second-year resident performance and could help decrease reading time and detection error.

Supranormal Left Ventricular Ejection Fraction, Stroke Volume, and Cardiovascular Risk: Findings From Population-Based Cohort Studies.

BACKGROUND: Supranormal ejection fraction by echocardiography in clinically referred patient populations has been associated with an increased risk of cardiovascular disease (CVD). The prognostic implication of supranormal left ventricular ejection fraction (LVEF)-assessed by cardiac magnetic resonance (CMR)-in healthy, community-dwelling individuals is unknown. OBJECTIVES: The purpose of this study is to investigate the prognostic implication of supranormal LVEF as assessed by CMR and its inter-relationship with stroke volume among community-dwelling adults without CVD. METHODS: Participants from the MESA (Multi-Ethnic Study of Atherosclerosis) and DHS (Dallas Heart Study) cohorts free of CVD who underwent CMR with LVEF above the normal CMR cutoff (≥57%) were included. The association between cohort-specific LVEF categories and risk of clinically adjudicated major adverse cardiovascular events (MACE) was assessed using adjusted Cox models. Subgroup analysis was also performed to evaluate the association of LVEF and risk of MACE among individuals stratified by left ventricular stroke volume index. RESULTS: The study included 4,703 participants from MESA and 2,287 from DHS with 727 and 151 MACE events, respectively. In adjusted Cox models, the risk of MACE was highest among individuals in LVEF Q4 (vs Q1) in both cohorts after accounting for potential confounders (MESA: HR = 1.27 [95% CI: 1.01-1.60], P = 0.04; DHS: HR = 1.72 [95% CI: 1.05-2.79], P = 0.03). A significant interaction was found between the continuous measures of LVEF and left ventricular stroke volume index (P interaction = 0.02) such that higher LVEF was significantly associated with an increased risk of MACE among individuals with low but not high stroke volume. CONCLUSIONS: Among community-dwelling adults without CVD, LVEF in the supranormal range is associated with a higher risk of adverse cardiovascular outcomes, particularly in those with lower stroke volume.

Deep Learning-Based COVID-19 Pneumonia Classification Using Chest CT Images: Model Generalizability.

Since the outbreak of the COVID-19 pandemic, worldwide research efforts have focused on using artificial intelligence (AI) technologies on various medical data of COVID-19-positive patients in order to identify or classify various aspects of the disease, with promising reported results. However, concerns have been raised over their generalizability, given the heterogeneous factors in training datasets. This study aims to examine the severity of this problem by evaluating deep learning (DL) classification models trained to identify COVID-19-positive patients on 3D computed tomography (CT) datasets from different countries. We collected one dataset at UT Southwestern (UTSW) and three external datasets from different countries: CC-CCII Dataset (China), COVID-CTset (Iran), and MosMedData (Russia). We divided the data into two classes: COVID-19-positive and COVID-19-negative patients. We trained nine identical DL-based classification models by using combinations of datasets with a 72% train, 8% validation, and 20% test data split. The models trained on a single dataset achieved accuracy/area under the receiver operating characteristic curve (AUC) values of 0.87/0.826 (UTSW), 0.97/0.988 (CC-CCCI), and 0.86/0.873 (COVID-CTset) when evaluated on their own dataset. The models trained on multiple datasets and evaluated on a test set from one of the datasets used for training performed better. However, the performance dropped close to an AUC of 0.5 (random guess) for all models when evaluated on a different dataset outside of its training datasets. Including MosMedData, which only contained positive labels, into the training datasets did not necessarily help the performance of other datasets. Multiple factors likely contributed to these results, such as patient demographics and differences in image acquisition or reconstruction, causing a data shift among different study cohorts.

Utilization of Structured Reporting to Monitor Outcomes of Doppler Ultrasound Performed for Deep Vein Thrombosis.

Determining the clinical impact of imaging exams at the enterprise level is problematic, as radiology reports historically have been created with the content meant primarily for the referring provider. Structured reporting can establish the foundation for enterprise monitoring of imaging outcomes without manual review providing the framework for assessment of utilization and quality. Ultrasound (US) for deep vein thrombosis evaluation (DVT) is an ideal testbed for assessing this functionality. The system standard template for Doppler US for extremity venous evaluation for DVT was updated with a discrete fixed picklist of impression options and implemented system wide. Template utilization and interpretive outcomes were actively monitored and use reinforced as part of standard clinical practice. From January 1, 2017 to December 31, 2017, 9111 US exams for DVT were performed with 8997 utilizing structured reporting (98.75%). Of those in the structured reporting group, 1074 (11.79%) were positive for any type of DVT with 732 (8.03%) reported as Acute/New above the knee. Positive rates for any type of DVT were 10.29% emergency department, 14.17% inpatient, and 13.20% outpatient. While being the lowest positive rate, the emergency department had the highest overall volume of exams. Structured reporting for DVT US assessment outcomes can be implemented with a very high rate of radiologist adoption and adherence providing accurate determination of positive rates, month by month, in differing patient locations. Structured elements can be used to automatically trigger downstream processes; in our institution, this will alert providers in the EHR if the patient does not receive anticoagulation within 2 h of a positive test. This lays the foundation for effective enterprise assessment of imaging outcomes forming the basis of future quality and safety initiatives on optimizing health system resource utilization.

Detrimental effect of systemic vascular risk factors on brain hemodynamic function assessed with MRI.

Background and purpose Vascular risk factors have been associated with decreased cerebral blood flow (CBF) but this is etiologically nonspecific and may result from vascular insufficiency or a response to decreased brain metabolic activity. We apply new MRI techniques to measure oxygen extraction fraction (OEF) and cerebral metabolic rate of oxygen consumption (CMRO2), hypothesizing that decreased CBF related to these vascular risk factors will be associated with increased OEF, confirming a primary vascular insufficiency. Methods 3T MRI was obtained on 70 community-based participants in this IRB-approved study with informed consent, with previous assessment of systolic blood pressure, hypertension medication, elevated serum triglycerides, low serum HDL, and diabetes mellitus. CBF was measured using phase contrast adjusted for brain volume (ml/100 g/min), OEF (%) was obtained from T2-Relaxation-Under-Spin-Tagging (TRUST), and CMRO2 (µmol/100 g/min) was derived using the Fick principle. Stepwise linear regression identified optimal predictors of CBF with age, sex, and hematocrit included for adjustment. This predictive model was then evaluated against OEF and CMRO2. Results Hypertriglyceridemia was associated with low CBF and high OEF. High systolic blood pressure was associated with high CBF and low OEF, which was primarily attributable to those with pressures above 160 mmHg. Neither risk factor was associated with significant differences in cerebral metabolic rate. Conclusion Low CBF related to hypertriglyceridemia was accompanied by high OEF with no significant difference in CMRO2, confirming subclinical vascular insufficiency. High CBF related to high systolic blood pressure likely reflected limitations of autoregulation at higher blood pressures.

Cognitive impact of lacunar infarcts and white matter hyperintensity volume.

BACKGROUND: Subcortical lacunar infarcts and white matter hyperintensities (WMH) are common neuroradiological findings, but few studies associate between these insults and cognition in a community-dwelling population. METHODS: The Dallas Heart Study is a population-based initiative whose assessments included demographic and clinical findings including brain MRI and the Montreal Cognitive Assessment (MoCA). The presence and number of lacunes in subjects aged over 55 years were assessed by study physicians. The WMH volume was measured by an automated method. The association between the presence and number of lacunar infarcts and of WMH volume with the total MoCA score and subdomains was assessed using linear regression with adjustment for age, gender and self-reported ethnicity. RESULTS: In 609 subjects with valid data, both the presence and the increasing number of lacunes were associated with lower MoCA scores, even after adjusting for demographic variables. The presence of lacunes was also associated with lower scores in the memory, executive and attention subdomains. The WMH volume was not significantly associated with the MoCA score. CONCLUSION: The presence and increasing number of lacunes in midlife is associated with a lower performance in multiple domains of a cognitive screening measure after adjusting for demographic factors.

Is coronary computed tomography angiography a resource sparing strategy in the risk stratification and evaluation of acute chest pain? Results of a randomized controlled trial.

OBJECTIVES: Annually, almost 6 million U.S. citizens are evaluated for acute chest pain syndromes (ACPSs), and billions of dollars in resources are utilized. A large part of the resource utilization results from precautionary hospitalizations that occur because care providers are unable to exclude the presence of coronary artery disease (CAD) as the underlying cause of ACPSs. The purpose of this study was to examine whether the addition of coronary computerized tomography angiography (CCTA) to the concurrent standard care (SC) during an index emergency department (ED) visit could lower resource utilization when evaluating for the presence of CAD. METHODS: Sixty participants were assigned randomly to SC or SC + CCTA groups. Participants were interviewed at the index ED visit and at 90 days. Data collected included demographics, perceptions of the value of accessing health care, and clinical outcomes. Resource utilization included services received from both the primary in-network and the primary out-of-network providers. The prospectively defined primary endpoint was the total amount of resources utilized over a 90-day follow-up period when adding CCTA to the SC risk stratification in ACPSs. RESULTS: The mean (± standard deviation [SD]) for total resources utilized at 90 days for in-network plus out-of-network services was less for the participants in the SC + CCTA group ($10,134; SD ±$14,239) versus the SC-only group ($16,579; SD ±$19,148; p = 0.144), as was the median for the SC + CCTA ($4,288) versus SC only ($12,148; p = 0.652; median difference = -$1,291; 95% confidence interval [CI] = -$12,219 to $1,100; p = 0.652). Among the 60 total study patients, only 19 had an established diagnosis of CAD at 90 days. However, 18 (95%) of these diagnosed participants were in the SC + CCTA group. In addition, there were fewer hospital readmissions in the SC + CCTA group (6 of 30 [20%] vs. 16 of 30 [53%]; difference in proportions = -33%; 95% CI = -56% to -10%; p = 0.007). CONCLUSIONS: Adding CCTA to the current ED risk stratification of ACPSs resulted in no difference in the quantity of resources utilized, but an increased diagnosis of CAD, and significantly less recidivism and rehospitalization over a 90-day follow-up period.

Changes in right ventricular structure and function assessed using cardiac magnetic resonance imaging in bosentan-treated patients with pulmonary arterial hypertension.

Patients with pulmonary arterial hypertension (PAH) usually show improvements in symptoms, exercise capacity, and hemodynamics after treatment with approved medical therapies. This study sought to determine whether improvement in right-sided cardiac function measured using cardiac magnetic resonance imaging would also be seen and whether these changes would correlate with improvement in exercise capacity. Sixteen patients with PAH underwent evaluation at baseline and after 12 months of treatment with bosentan. After treatment, cardiac index, pulmonary vascular resistance, and 6-minute walk distance improved, and there was a trend toward improvement in right ventricular (RV) stroke volume (70 +/- 27 to 81 +/- 30 ml; p = 0.08), but no change in RV ejection fraction (RVEF) or RV end-diastolic volume. Six-minute walk distance improved by 59 m (p <0.05) in the overall cohort and improved more in patients in whom RVEF increased compared with those with stable or decreased RVEF (+98 vs -37 m, respectively; p = 0.01). Three patients died during follow-up, and these patients had significantly lower RVEF and left ventricular end-diastolic volume indexes than surviving patients. In conclusion, these results suggest that cardiac magnetic resonance imaging may have value in determining response to therapy and prognosis in patients with PAH.

Noninvasive cardiac imaging in pulmonary hypertension.

Pulmonary artery hypertension is a rare disease with significant morbidity and mortality. Initial and serial noninvasive assessment of these patients can be accomplished with transthoracic echocardiography and/or cardiac magnetic resonance imaging. These complementary techniques provide the structural and functional information required to care for patients with pulmonary artery hypertension and are discussed in this review.

Influence of body fat content and distribution on variation in metabolic risk.

OBJECTIVES: Several reports indicate that the body fat compartments, especially ip fat, predict metabolic risk better than total body fat. The objective of the study was to determine whether this can be confirmed and generalized throughout the population. PARTICIPANTS: A representative sample of 1934 Black and White women and men of the Dallas Heart Study participated in the study. DESIGN: We measured the fat in total body, trunk, and lower body with dual-energy x-ray absorptiometry and in abdominal compartments (sc, ip, and retroperitoneal) with magnetic resonance imaging. Other measurements included body mass index (BMI), waist circumference, blood pressure, plasma lipids, glucose, insulin (including homeostasis model), and C-reactive protein. RESULTS: In all groups, total body fat correlated positively with key metabolic risk factors, i.e. homeostasis model, triglyceride/high-density lipoprotein-cholesterol ratios, C-reactive protein, and blood pressure; however, it explained less than one third of the variability of all the risk factors. After adjustment for total body fat, truncal fat conferred additional positive correlation with risk factors. Furthermore, with multivariable regression analysis, ip fat conferred independent correlation with plasma lipids beyond a combination of other compartments including truncal fat. Still, except for insulin levels, all combinations including ip fat still explained less than one third of the variability in risk-factor levels. Conversely, lower body fat correlated negatively with risk factors; i.e. lower body fat appeared to offer some protection against risk factors. CONCLUSIONS: Body fat distribution has some influence on risk factors beyond total body fat content. Both waist circumference and BMI significantly predicted risk factors after adjustment for total body fat, and for clinical purposes, most of the predictive power for men was contained in waist circumference, whereas for women, BMI and waist circumference were similarly predictive. Finally, even though the correlations between combined body fat parameters and risk factors explained only a portion of the variation in the latter, the average number of categorical metabolic risk factors increased progressively with increasing obesity. Hence, obesity seemingly has more clinical impact than revealed in these correlative studies.