CNNs trained with adult data are useful in pediatrics. A pneumonia classification example.

BACKGROUND AND OBJECTIVES: The scarcity of data for training deep learning models in pediatrics has prompted questions about the feasibility of employing CNNs trained with adult images for pediatric populations. In this work, a pneumonia classification CNN was used as an exploratory example to showcase the adaptability and efficacy of such models in pediatric healthcare settings despite the inherent data constraints. METHODS: To develop a curated training dataset with reduced biases, 46,947 chest X-ray images from various adult datasets were meticulously selected. Two preprocessing approaches were tried to assess the impact of thoracic segmentation on model attention outside the thoracic area. Evaluation of our approach was carried out on a dataset containing 5,856 chest X-rays of children from 1 to 5 years old. RESULTS: An analysis of attention maps indicated that networks trained with thorax segmentation placed less attention on regions outside the thorax, thus eliminating potential bias. The ensuing network exhibited impressive performance when evaluated on an adult dataset, achieving a pneumonia discrimination AUC of 0.95. When tested on a pediatric dataset, the pneumonia discrimination AUC reached 0.82. CONCLUSIONS: The results of this study show that adult-trained CNNs can be effectively applied to pediatric populations. This could potentially shift focus towards validating adult models over pediatric population instead of training new CNNs with limited pediatric data. To ensure the generalizability of deep learning models, it is important to implement techniques aimed at minimizing biases, such as image segmentation or low-quality image exclusion.

Quantitative Interstitial Abnormality Progression and Outcomes in the Genetic Epidemiology of COPD and Pittsburgh Lung Screening Study Cohorts.

BACKGROUND: The risk factors and clinical outcomes of quantitative interstitial abnormality progression over time have not been characterized. RESEARCH QUESTIONS: What are the associations of quantitative interstitial abnormality progression with lung function, exercise capacity, and mortality? What are the demographic and genetic risk factors for quantitative interstitial abnormality progression? STUDY DESIGN AND METHODS: Quantitative interstitial abnormality progression between visits 1 and 2 was assessed from 4,635 participants in the Genetic Epidemiology of COPD (COPDGene) cohort and 1,307 participants in the Pittsburgh Lung Screening Study (PLuSS) cohort. We used multivariable linear regression to determine the risk factors for progression and the longitudinal associations between progression and FVC and 6-min walk distance, and Cox regression models for the association with mortality. RESULTS: Age at enrollment, female sex, current smoking status, and the MUC5B minor allele were associated with quantitative interstitial abnormality progression. Each percent annual increase in quantitative interstitial abnormalities was associated with annual declines in FVC (COPDGene: 8.5 mL/y; 95% CI, 4.7-12.4 mL/y; P < .001; PLuSS: 9.5 mL/y; 95% CI, 3.7-15.4 mL/y; P = .001) and 6-min walk distance, and increased mortality (COPDGene: hazard ratio, 1.69; 95% CI, 1.34-2.12; P < .001; PLuSS: hazard ratio, 1.28; 95% CI, 1.10-1.49; P = .001). INTERPRETATION: The objective, longitudinal measurement of quantitative interstitial abnormalities may help identify people at greatest risk for adverse events and most likely to benefit from early intervention.

Quantification of Arterial and Venous Morphologic Markers in Pulmonary Arterial Hypertension Using CT Imaging.

BACKGROUND: Pulmonary hypertension is a heterogeneous disease, and a significant portion of patients at risk for it have CT imaging available. Advanced automated processing techniques could be leveraged for early detection, screening, and development of quantitative phenotypes. Pruning and vascular tortuosity have been previously described in pulmonary arterial hypertension (PAH), but the extent of these phenomena in arterial vs venous pulmonary vasculature and in exercise pulmonary hypertension (ePH) have not been described. RESEARCH QUESTION: What are the arterial and venous manifestations of pruning and vascular tortuosity using CT imaging in PAH, and do they also occur in ePH? STUDY DESIGN AND METHODS: A cohort of patients with PAH and ePH and control subjects with available CT angiograms were retrospectively identified to examine the differential arterial and venous presence of pruning and tortuosity in patients with precapillary pulmonary hypertension not confounded by lung or thromboembolic disease. The pulmonary vasculature was reconstructed, and an artificial intelligence method was used to separate arteries and veins and to compute arterial and venous vascular volumes and tortuosity. RESULTS: A total of 42 patients with PAH, 12 patients with ePH, and 37 control subjects were identified. There was relatively lower (median [interquartile range]) arterial small vessel volume in subjects with PAH (PAH 14.7 [11.7-16.5; P < .0001]) vs control subjects (16.9 [15.6-19.2]) and venous small vessel volume in subjects with PAH and ePH (PAH 8.0 [6.5-9.6; P < .0001]; ePH, 7.8 [7.5-11.4; P = .004]) vs control subjects (11.5 [10.6-12.2]). Higher large arterial volume, however, was only observed in the pulmonary arteries (PAH 17.1 [13.6-23.4; P < .0001] vs control subjects 11.4 [8.1-15.4]). Similarly, tortuosity was higher in the pulmonary arteries in the PAH group (PAH 3.5 [3.3-3.6; P = .0002] vs control 3.2 [3.2-3.3]). INTERPRETATION: Lower small distal pulmonary vascular volume, higher proximal arterial volume, and higher arterial tortuosity were observed in PAH. These can be quantified by using automated techniques from clinically acquired CT scans of patients with ePH and resting PAH.

Loss of Pulmonary Vascular Volume as a Predictor of Right Ventricular Dysfunction and Mortality in Acute Pulmonary Embolism.

BACKGROUND: In acute pulmonary embolism, chest computed tomography angiography derived metrics, such as the right ventricle (RV): left ventricle ratio are routinely used for risk stratification. Paucity of intraparenchymal blood vessels has previously been described, but their association with clinical biomarkers and outcomes has not been studied. We sought to determine if small vascular volumes measured on computed tomography scans were associated with an abnormal RV on echocardiography and mortality. We hypothesized that decreased small venous volume would be associated with greater RV dysfunction and increased mortality. METHODS: A retrospective cohort of patients with intermediate risk pulmonary embolism admitted to Brigham and Women's Hospital between 2009 and 2017 was assembled, and clinical and radiographic data were obtained. We performed 3-dimensional reconstructions of vasculature to assess intraparenchymal vascular volumes. Statistical analyses were performed using multivariable regression and cox proportional hazards models, adjusting for age, sex, lung volume, and small arterial volume. RESULTS: Seven hundred twenty-two subjects were identified of whom 573 had documented echocardiography. A 50% reduction in small venous volume was associated with an increased risk of RV dilation (relative risk: 1.38 [95% CI, 1.18-1.63], P<0.001), RV dysfunction (relative risk: 1.62 [95% CI, 1.36-1.95], P<0.001), and RV strain (relative risk: 1.67 [95% CI, 1.37-2.04], P<0.001); increased cardiac biomarkers, and higher 30-day and 90-day mortality (hazard ratio: 2.50 [95% CI, 1.33-4.67], P=0.004 and hazard ratio: 1.84 [95% CI, 1.11-3.04], P=0.019, respectively). CONCLUSIONS: Loss of small venous volume quantified from computed tomography angiography is associated with increased risk of abnormal RV on echocardiography, abnormal cardiac biomarkers, and higher risk of 30- and 90-day mortality. Small venous volume may be a useful marker for assessing disease severity in acute pulmonary embolism.

Estimated Ventricular Size, Asthma Severity, and Exacerbations: The Severe Asthma Research Program III Cohort.

BACKGROUND: Relative enlargement of the pulmonary artery (PA) on chest CT imaging is associated with respiratory exacerbations in patients with COPD or cystic fibrosis. We sought to determine whether similar findings were present in patients with asthma and whether these findings were explained by differences in ventricular size. METHODS: We measured the PA and aorta diameters in 233 individuals from the Severe Asthma Research Program III cohort. We also estimated right, left, and total epicardial cardiac ventricular volume indices (eERVVI, eELVVI, and eETVVI, respectively). Associations between the cardiac and PA measures (PA-to-aorta [PA/A] ratio, eERVVI-to-eELVVI [eRV/eLV] ratio, eERVVI, eELVVI, eETVVI) and clinical measures of asthma severity were assessed by Pearson correlation, and associations with asthma severity and exacerbation rate were evaluated by multivariable linear and zero-inflated negative binomial regression. RESULTS: Asthma severity was associated with smaller ventricular volumes. For example, those with severe asthma had 36.1 mL/m2 smaller eETVVI than healthy control subjects (P = .003) and 14.1 mL/m2 smaller eETVVI than those with mild/moderate disease (P = .011). Smaller ventricular volumes were also associated with a higher rate of asthma exacerbations, both retrospectively and prospectively. For example, those with an eETVVI less than the median had a 57% higher rate of exacerbations during follow-up than those with eETVVI greater than the median (P = .020). Neither PA/A nor eRV/eLV was associated with asthma severity or exacerbations. CONCLUSIONS: In patients with asthma, smaller cardiac ventricular size may be associated with more severe disease and a higher rate of asthma exacerbations. TRIAL REGISTRY: ClinicalTrials.gov; No.: NCT01761630; URL: www.clinicaltrials.gov.

Atlas-based segmentation of white matter structures from DTI using tensor invariants and orientation.

This paper presents a novel method for the segmentation of anatomical structures in the white matter from DTI (Diffusion Tensor Imaging) data. Our approach is based on: (a) the use of a DTI white matter atlas to guide the segmentation process, (b) the use of tensor invariants and the orientation information of the tensor as features, and (c) a statistical modeling of the data with a level set implementation. This formulation allows for controlling the relative importance of the different properties of the diffusion tensor and uses the anatomical information of the atlas to constrain the segmentation. The method has been applied to the segmentation of DTI volumes, and results show it constitutes a valid alternative to other approaches such as VBM or TBSS for white matter analysis.