Influence of augmentation on the performance of the double ResNet-based model for chest X-ray classification.

Purpose: A pandemic disease elicited by the SARS-CoV-2 virus has become a serious health issue due to infecting millions of people all over the world. Recent publications prove that artificial intelligence (AI) can be used for medical diagnosis purposes, including interpretation of X-ray images. X-ray scanning is relatively cheap, and scan processing is not computationally demanding. Material and methods: In our experiment a baseline transfer learning schema of processing of lung X-ray images, including augmentation, in order to detect COVID-19 symptoms was implemented. Seven different scenarios of augmentation were proposed. The model was trained on a dataset consisting of more than 30,000 X-ray images. Results: The obtained model was evaluated using real images from a Polish hospital, with the use of standard metrics, and it achieved accuracy = 0.9839, precision = 0.9697, recall = 1.0000, and F1-score = 0.9846. Conclusions: Our experiment proved that augmentations and masking could be important steps of data pre-processing and could contribute to improvement of the evaluation metrics. Because medical professionals often tend to lack confidence in AI-based tools, we have designed the proposed model so that its results would be explainable and could play a supporting role for radiology specialists in their work.

Supporting the Diagnosis of Fabry Disease Using a Natural Language Processing-Based Approach.

In clinical practice, the consideration of non-specific symptoms of rare diseases in order to make a correct and timely diagnosis is often challenging. To support physicians, we developed a decision-support scoring system on the basis of retrospective research. Based on the literature and expert knowledge, we identified clinical features typical for Fabry disease (FD). Natural language processing (NLP) was used to evaluate patients' electronic health records (EHRs) to obtain detailed information about FD-specific patient characteristics. The NLP-determined elements, laboratory test results, and ICD-10 codes were transformed and grouped into pre-defined FD-specific clinical features that were scored in the context of their significance in the FD signs. The sum of clinical feature scores constituted the FD risk score. Then, medical records of patients with the highest FD risk score were reviewed by physicians who decided whether to refer a patient for additional tests or not. One patient who obtained a high-FD risk score was referred for DBS assay and confirmed to have FD. The presented NLP-based, decision-support scoring system achieved AUC of 0.998, which demonstrates that the applied approach enables for accurate identification of FD-suspected patients, with a high discrimination power.

Leukocyte deep learning classification assessment using Shapley additive explanations algorithm.

INTRODUCTION: A peripheral blood smear is a basic test for hematological disease diagnosis. This test is performed manually in many places worldwide, which requires both time and qualified staff. Large laboratories are equipped with digital morphology analyzers, some of which are based on deep learning methods. However, it is difficult to explain to scientists how they work. In this paper, we proposed to add an explanatory factor to enhance the interpretability of deep learning models in leukocyte classification. METHODS: 10 297 single images of leukocytes obtained from peripheral blood smears were included in this study. Pre-trained and fully trained VGG16 and VGG19 models were used to classify the leukocytes, and Shapley Additive Explanations (SHAP) DeepExplainer was applied to visualize the area of cells that were significant for classification. The output images from the DeepExplainer were compared with cellular elements that are essential to laboratory practice. RESULTS: The accuracy of our fully trained models was 99.81% for VGG16 and 99.79% for VGG19. It achieved slightly better results than the partially trained model, which scored 98.67% for VGG16 and 98.33% for VGG19. Their SHAP explanations indicated the significance of cellular structures in microscopic examination. Explanations in the pre-trained models have proved the cell and nucleus contours to be relevant to classification, while explanations in the fully trained models pointed to the cytoplasm area. CONCLUSION: Despite different SHAP DeepExplainer explanations for fully and partially trained models, this method appears to be helpful for the verification of leukocyte classification in automated peripheral blood smear examination.

Prepregnancy exposure to dietary arsenic and congenital heart defects.

INTRODUCTION: Arsenic crosses the placenta and accumulates in fetal tissues. In the United States, diet is the predominant route of arsenic exposure, but epidemiologic data are sparse regarding this exposure and development of birth defects. Using data from a large case-control study, we explored associations between maternal dietary arsenic exposure and congenital heart defects (CHDs), the most prevalent birth defects. METHODS: We used maternal self-reported dietary assessments and arsenic concentration estimates in food items to estimate average daily exposure to dietary arsenic during the year before pregnancy for mothers of 10,446 unaffected control children and 6,483 case children diagnosed with CHDs. Using tertiles of dietary exposure to total arsenic (all species) and inorganic arsenic, we applied logistic regression analysis to estimate associations for middle and high tertiles, compared with the low tertile. RESULTS: Positive associations (odds ratio [OR] ≥ 1.2) for total arsenic were observed in both tertiles for perimembranous ventricular septal defect (VSD) and high tertile only for double outlet right ventricle-transposition of the great arteries (DORV-TGA), partial anomalous pulmonary venous return (PAPVR), and tricuspid atresia. Positive associations were also observed in both tertiles (tricuspid atresia) and high tertile only (DORV-TGA, conoventricular VSD, PAPVR, and pulmonary atresia) for inorganic arsenic. Most remaining associations were near or below unity. DISCUSSION: Exploration of maternal dietary exposure to total and inorganic arsenic and CHDs produced few positive associations but was limited by available food item concentrations. Future research requires expanded collection of dietary data, improved estimates of concentrations, and consideration of nondietary sources of arsenic exposure.

Pre-pregnancy exposure to arsenic in diet and non-cardiac birth defects.

OBJECTIVES: To explore associations between maternal pre-pregnancy exposure to arsenic in diet and non-cardiac birth defects. DESIGN: This is a population-based, case-control study using maternal responses to a dietary assessment and published arsenic concentration estimates in food items to calculate average daily total and inorganic arsenic exposure during the year before pregnancy. Assigning tertiles of total and inorganic arsenic exposure, logistic regression analysis was used to estimate OR for middle and high tertiles, compared to the low tertile. SETTING: US National Birth Defects Prevention Study, 1997-2011. PARTICIPANTS: Mothers of 10 446 children without birth defects and 14 408 children diagnosed with a non-cardiac birth defect. RESULTS: Maternal exposure to total dietary arsenic in the middle and high tertiles was associated with a threefold increase in cloacal exstrophy, with weak positive associations (1·2-1·5) observed either in both tertiles (intercalary limb deficiency) or the high tertile only (encephalocele, glaucoma/anterior chamber defects and bladder exstrophy). Maternal exposure to inorganic arsenic showed mostly weak, positive associations in both tertiles (colonic atresia/stenosis, oesophageal atresia, bilateral renal agenesis/hypoplasia, hypospadias, cloacal exstrophy and gastroschisis), or the high (glaucoma/anterior chamber defects, choanal atresia and intestinal atresia stenosis) or middle (encephalocele, intercalary limb deficiency and transverse limb deficiency) tertiles only. The remaining associations estimated were near the null or inverse. CONCLUSIONS: This exploration of arsenic in diet and non-cardiac birth defects produced several positive, but mostly weak associations. Limitations in exposure assessment may have resulted in exposure misclassification. Continued research with improved exposure assessment is recommended to identify if these associations are true signals or chance findings.

Is maternal employment site a source of exposure misclassification in studies of environmental exposures and birth outcomes? A simulation-based bias analysis of haloacetic acids in tap water and hypospadias.

In population research, exposure to environmental contaminants is often indirectly assessed by linking residence to geocoded databases of environmental exposures. We explored the potential for misclassification of residence-based environmental exposure as a result of not accounting for the workplace environments of employed pregnant women using data from a National Birth Defects Prevention Study (NBDPS) analysis of drinking water haloacetic acids and hypospadias. Methods: The original analysis used NBDPS data from women with haloacetic acid exposure information in eight states who delivered an infant with second- or third-degree hypospadias (cases) or a male infant without a birth defect (controls) between 2000 and 2005. In this bias analysis, we used a uniform distribution to randomly select 11%-14% of employed women that were assumed to change municipal water systems between home and work and imputed new contaminant exposures for tap water beverages consumed at work among the selected women using resampled values from the control population. Multivariable logistic regression was used to estimate the association between hypospadias and haloacetic acid ingestion with the same covariates and exposure cut-points as the original study. We repeated this process across 10,000 iterations and then completed a sensitivity analysis of an additional 10,000 iterations where we expanded the uniform distribution (i.e., 0%, 28%). Results: In both simulations, the average results of the 10,000 iterations were nearly identical to those of the initial study. Conclusions: Our results suggest that household estimates may be sufficient proxies for worksite exposures to haloacetic acids in tap water.

Maternal Exposure to Disinfection By-Products and Risk of Hypospadias in the National Birth Defects Prevention Study (2000-2005).

The purpose of this study was to estimate the association between 2nd and 3rd degree hypospadias and maternal exposure to disinfection by-products (DBPs) using data from a large case-control study in the United States. Concentration estimates for total trihalomethanes (TTHMs), the sum of the five most prevalent haloacetic acids (HAA5), and individual species of each were integrated with data on maternal behaviors related to water use from the National Birth Defects Prevention Study (NBDPS) to create three different exposure metrics: (1) household DBP concentrations; (2) estimates of DBP ingestion; (3) predicted uptake (i.e., internal dose) of trihalomethanes (THMs) via ingestion, showering, and bathing. The distribution of DBP exposure was categorized as follows: (Q1/referent) < 50%; (Q2) ≥ 50% to < 75%; and (Q3) ≥ 75%. Logistic regression was used to estimate adjusted odds ratios (aORs) and 95% confidence intervals (CIs). Generally, null associations were observed with increasing TTHM or HAA5 exposure. An increased risk was observed among women with household bromodichloromethane levels in the second quantile (aOR: 1.8; 95% CI: 1.2, 2.7); however, this association did not persist after the inclusion of individual-level water-use data. Findings from the present study do not support the hypothesis that maternal DBP exposures are related to the occurrence of hypospadias.

Pre-pregnancy dietary arsenic consumption among women in the United States.

BACKGROUND: Arsenic is associated with several adverse health outcomes, including some birth defects. Although diet is the predominant route of arsenic exposure in the United States (U.S.), limited data exist regarding pre-pregnancy dietary arsenic consumption among U.S. women. METHODS: Using data collected in the National Birth Defects Prevention Study (NBDPS), we estimated daily dietary arsenic consumption during the year before pregnancy for 10,886 mothers of nonmalformed control children delivered from 1997-2011. Responses to the NBDPS dietary assessment and food item estimates of total and inorganic arsenic were used to estimate consumption. Associations between total and inorganic arsenic consumption and selected maternal characteristics were estimated using multinomial logistic regression. RESULTS: Estimates of mean maternal total and inorganic dietary arsenic consumption were 14.9 and 5.2 µg/day, respectively. Several positive and inverse associations with confidence intervals that excluded the null were observed. Comparing mothers in the middle or high total arsenic consumption tertiles to those in the low tertile, we observed positive associations (odds ratios = 1.3-3.8) for maternal age (≥30 years), lower (0-8 years) or higher (>12 years) education, race/ethnicity (non-Hispanic Black, Hispanic, other), and early pregnancy drinking with no binge episodes, and inverse associations (odds ratios = 0.4-0.8) for age (<25 years), body mass index (≥30.0 kg/m2 ), and early pregnancy smoking. Findings tended to be similar for inorganic arsenic consumption. CONCLUSIONS: These contemporary estimates of pre-pregnancy dietary arsenic consumption among U.S. women show associations between both total and inorganic dietary arsenic consumption and several maternal characteristics, improving characterization of the public health impact of this exposure.

Intensive trapping of blood-fed Anopheles darlingi in Amazonian Peru reveals unexpectedly high proportions of avian blood-meals.

Anopheles darlingi, the main malaria vector in the Neotropics, has been considered to be highly anthropophilic. However, many behavioral aspects of this species remain unknown, such as the range of blood-meal sources. Barrier screens were used to collect resting Anopheles darlingi mosquitoes from 2013 to 2015 in three riverine localities (Lupuna, Cahuide and Santa Emilia) in Amazonian Peru. Overall, the Human Blood Index (HBI) ranged from 0.58-0.87, with no significant variation among years or sites. Blood-meal analysis revealed that humans are the most common blood source, followed by avian hosts (Galliformes-chickens and turkeys), and human/Galliforme mixed-meals. The Forage Ratio and Selection Index both show a strong preference for Galliformes over humans in blood-fed mosquitoes. Our data show that 30% of An. darlingi fed on more than one host, including combinations of dogs, pigs, goats and rats. There appears to be a pattern of host choice in An. darlingi, with varying proportions of mosquitoes feeding only on humans, only on Galliformes and some taking mixed-meals of blood (human plus Galliforme), which was detected in the three sites in different years, indicating that there could be a structure to these populations based on blood-feeding preferences. Mosquito age, estimated in two localities, Lupuna and Cahuide, ranged widely between sites and years. This variation may reflect the range of local environmental factors that influence longevity or possibly potential changes in the ability of the mosquito to transmit the parasite. Of 6,204 resting An. darlingi tested for Plasmodium infection, 0.42% were infected with P. vivax. This study provides evidence for the first time of the usefulness of barrier screens for the collection of blood-fed resting mosquitoes to calculate the Human Blood Index (HBI) and other blood-meal sources in a neotropical malaria endemic setting.

Paternal and joint parental occupational pesticide exposure and spina bifida in the National Birth Defects Prevention Study, 1997 to 2002.

BACKGROUND: Because of persistent concerns over the association between pesticides and spina bifida, we examined the role of paternal and combined parental occupational pesticide exposures in spina bifida in offspring using data from a large population-based study of birth defects. METHODS: Occupational information from fathers of 291 spina bifida cases and 2745 unaffected live born control infants with estimated dates of delivery from 1997 to 2002 were collected by means of maternal report. Two expert industrial hygienists estimated exposure intensity and frequency to insecticides, herbicides, and fungicides. Multivariable logistic regression models were used to estimate adjusted odds ratios (aOR) and 95% confidence intervals (CI) for exposure to any pesticide and to any class of pesticide (yes/no; and by median), and exposure to combinations of pesticides (yes/no) and risk of spina bifida. Adjusted odds ratios were also estimated by parent exposed to pesticides (neither, mother only, father only, both parents). RESULTS: Joint parental occupational pesticide exposure was positively associated with spina bifida (aOR, 1.5; 95% CI, 0.9-2.4) when compared with infants with neither maternal nor paternal exposures; a similar association was not observed when only one parent was exposed. There was a suggested positive association between combined paternal insecticide and fungicide exposures and spina bifida (aOR, 1.5; 95% CI, 0.8-2.8), however, nearly all other aORs were close to unity. CONCLUSION: Overall, there was little evidence paternal occupational pesticide exposure was associated with spina bifida. However, the small numbers make it difficult to precisely evaluate the role of pesticide classes, individually and in combination. Birth Defects Research (Part A) 106:963-971, 2016. © 2016 Wiley Periodicals, Inc.

Birth defects in infants born to employees of a microelectronics and business machine manufacturing facility.

BACKGROUND: Concerns about solvent releases from a microelectronics/business machine manufacturing facility in upstate New York led to interest in the health of former workers, including this investigation of birth defects in children of male and female employees. METHODS: Children born 1983 to 2001 to facility employees were enumerated and matched to New York State's Congenital Malformations Registry. Reported structural birth defects were compared with numbers expected from state rates (excluding New York City), generating standardized prevalence ratios (SPRs). Exposure assessors classified employees as ever/never potentially exposed at the facility to metals, chlorinated hydrocarbons, and other hydrocarbons during windows critical to organogenesis (female workers) or spermatogenesis (male workers). Among workers, adjusted prevalence ratios were generated to evaluate associations between potential exposures and specific birth defects. RESULTS: External comparisons for structural defects were at expectation for infants of male workers (SPR = 1.01; 95% confidence interval [CI], 0.77-1.29; n = 60) and lower for births to female workers (SPR = 0.84; 95% CI, 0.50-1.33; n = 18). Among full-term infants of male workers, ventricular septal defects (VSDs) were somewhat elevated compared with the general population (SPR = 1.58; 95% CI, 0.99-2.39; n = 22). Within the cohort, potential paternal metal exposure was associated with increased VSD risk (adjusted prevalence ratio = 2.70; 95% CI, = 1.09-6.67; n = 7). CONCLUSION: While overall SPRs were near expectation, paternal exposure to metals (primarily lead) appeared to be associated with increased VSD risk in infants. Take-home of occupational exposures, nonoccupational exposures, and chance could not be ruled out as causes. Case numbers for many defects were small, limiting assessment of the role of occupational exposures. Birth Defects Research (Part A) 106:696-707, 2016. © 2016 Wiley Periodicals, Inc.

Sex ratios among infants with birth defects, National Birth Defects Prevention Study, 1997-2009.

A small number of population-based studies have examined sex differences among infants with birth defects. This study presents estimates of sex ratio for both isolated cases and those with multiple congenital anomalies, as well as by race/ethnicity. Male-female sex ratios and their 95% confidence intervals were calculated for 25,952 clinically reviewed case infants included in the National Birth Defects Prevention Study (1997-2009), a large population-based case-control study of birth defects. The highest elevations in sex ratios (i.e., male preponderance) among isolated non-cardiac defects were for craniosynostosis (2.12), cleft lip with cleft palate (2.01), and cleft lip without cleft palate (1.78); the lowest sex ratios (female preponderance) were for choanal atresia (0.45), cloacal exstrophy (0.46), and holoprosencephaly (0.64). Among isolated cardiac defects, the highest sex ratios were for aortic stenosis (2.88), coarctation of the aorta (2.51), and d-transposition of the great arteries (2.34); the lowest were multiple ventricular septal defects (0.52), truncus arteriosus (0.63), and heterotaxia with congenital heart defect (0.64). Differences were observed by race/ethnicity for some but not for most types of birth defects. The sex differences we observed for specific defects, between those with isolated versus multiple defects, as well as by race/ethnicity, demonstrate patterns that may suggest etiology and improve classification.