Deep Neural Network-Aided Histopathological Analysis of Myocardial Injury

Deep neural networks have become the mainstream approach for analyzing and interpreting histology images. In this study, we established and validated an interpretable DNN model to assess endomyocardial biopsy (EMB) data of patients with myocardial injury. Deep learning models were used to extract features and classify EMB histopathological images of heart failure cases diagnosed with either ischemic cardiomyopathy or idiopathic dilated cardiomyopathy and non-failing cases (organ donors without a history of heart failure). We utilized the gradient-weighted class activation mapping (Grad-CAM) technique to emphasize injured regions, providing an entry point to assess the dominant morphology in the process of a comprehensive evaluation. To visualize clustered regions of interest (ROI), we utilized uniform manifold approximation and projection (UMAP) embedding for dimension reduction. We further implemented a multi-model ensemble mechanism to improve the quantitative metric (area under the receiver operating characteristic curve, AUC) to 0.985 and 0.992 on ROI-level and case-level, respectively, outperforming the achievement of 0.971 ± 0.017 and 0.981 ± 0.020 based on the sub-models. Collectively, this new methodology provides a robust and interpretive framework to explore local histopathological patterns, facilitating the automatic and high-throughput quantification of cardiac EMB analysis.

Download Full-text

Evaluation of factors that predict the success rate of trial of labor after the cesarean section

BMC Pregnancy and Childbirth ◽

10.1186/s12884-021-04004-z ◽

2021 ◽

Vol 21 (1) ◽

Author(s):

Yang Mi ◽

Pengfei Qu ◽

Na Guo ◽

Ruimiao Bai ◽

Jiayi Gao ◽

...

Keyword(s):

Logistic Regression ◽

Cesarean Section ◽

Receiver Operating Characteristic Curve ◽

Success Rate ◽

Operating Characteristic ◽

Characteristic Curve ◽

Predictive Ability ◽

Training Set ◽

History Of ◽

Operating Characteristic Curve

Abstract Background For most women who have had a previous cesarean section, vaginal birth after cesarean section (VBAC) is a reasonable and safe choice, but which will increase the risk of adverse outcomes such as uterine rupture. In order to reduce the risk, we evaluated the factors that may affect VBAC and and established a model for predicting the success rate of trial of the labor after cesarean section (TOLAC). Methods All patients who gave birth at Northwest Women’s and Children’s Hospital from January 2016 to December 2018, had a history of cesarean section and voluntarily chose the TOLAC were recruited. Among them, 80% of the population was randomly assigned to the training set, while the remaining 20% were assigned to the external validation set. In the training set, univariate and multivariate logistic regression models were used to identify indicators related to successful TOLAC. A nomogram was constructed based on the results of multiple logistic regression analysis, and the selected variables included in the nomogram were used to predict the probability of successfully obtaining TOLAC. The area under the receiver operating characteristic curve was used to judge the predictive ability of the model. Results A total of 778 pregnant women were included in this study. Among them, 595 (76.48%) successfully underwent TOLAC, whereas 183 (23.52%) failed and switched to cesarean section. In multi-factor logistic regression, parity = 1, pre-pregnancy BMI < 24 kg/m2, cervical score ≥ 5, a history of previous vaginal delivery and neonatal birthweight < 3300 g were associated with the success of TOLAC. The area under the receiver operating characteristic curve in the prediction and validation models was 0.815 (95% CI: 0.762–0.854) and 0.730 (95% CI: 0.652–0.808), respectively, indicating that the nomogram prediction model had medium discriminative power. Conclusion The TOLAC was useful to reducing the cesarean section rate. Being primiparous, not overweight or obese, having a cervical score ≥ 5, a history of previous vaginal delivery or neonatal birthweight < 3300 g were protective indicators. In this study, the validated model had an approving predictive ability.

Download Full-text

Traditional Cardiovascular Risk Factors and Individual Prediction of Cardiovascular Events in Childhood Cancer Survivors

JNCI Journal of the National Cancer Institute ◽

10.1093/jnci/djz108 ◽

2019 ◽

Vol 112 (3) ◽

pp. 256-265 ◽

Cited By ~ 6

Author(s):

Yan Chen ◽

Eric J Chow ◽

Kevin C Oeffinger ◽

William L Border ◽

Wendy M Leisenring ◽

...

Keyword(s):

Risk Factors ◽

Heart Failure ◽

Cardiovascular Risk ◽

Childhood Cancer ◽

Cardiovascular Risk Factors ◽

Operating Characteristic ◽

Prediction Models ◽

Characteristic Curve ◽

Childhood Cancer Survivors ◽

Operating Characteristic Curve

Abstract Background Childhood cancer survivors have an increased risk of heart failure, ischemic heart disease, and stroke. They may benefit from prediction models that account for cardiotoxic cancer treatment exposures combined with information on traditional cardiovascular risk factors such as hypertension, dyslipidemia, and diabetes. Methods Childhood Cancer Survivor Study participants (n = 22 643) were followed through age 50 years for incident heart failure, ischemic heart disease, and stroke. Siblings (n = 5056) served as a comparator. Participants were assessed longitudinally for hypertension, dyslipidemia, and diabetes based on self-reported prescription medication use. Half the cohort was used for discovery; the remainder for replication. Models for each outcome were created for survivors ages 20, 25, 30, and 35 years at the time of prediction (n = 12 models). Results For discovery, risk scores based on demographic, cancer treatment, hypertension, dyslipidemia, and diabetes information achieved areas under the receiver operating characteristic curve and concordance statistics 0.70 or greater in 9 and 10 of the 12 models, respectively. For replication, achieved areas under the receiver operating characteristic curve and concordance statistics 0.70 or greater were observed in 7 and 9 of the models, respectively. Across outcomes, the most influential exposures were anthracycline chemotherapy, radiotherapy, diabetes, and hypertension. Survivors were then assigned to statistically distinct risk groups corresponding to cumulative incidences at age 50 years of each target outcome of less than 3% (moderate-risk) or approximately 10% or greater (high-risk). Cumulative incidence of all outcomes was 1% or less among siblings. Conclusions Traditional cardiovascular risk factors remain important for predicting risk of cardiovascular disease among adult-age survivors of childhood cancer. These prediction models provide a framework on which to base future surveillance strategies and interventions.

Download Full-text

Early postpartum dyslipidemia and its potential predictors during pregnancy in women with a history of gestational diabetes mellitus

10.21203/rs.3.rs-52856/v2 ◽

2020 ◽

Author(s):

Ling Pei ◽

Huangmeng Xiao ◽

Fenghua Lai ◽

Zeting Li ◽

Zhuyu Li ◽

...

Keyword(s):

Diabetes Mellitus ◽

Gestational Diabetes ◽

Gestational Diabetes Mellitus ◽

Maternal Age ◽

Operating Characteristic ◽

Characteristic Curve ◽

Lipid Levels ◽

Early Postpartum ◽

History Of ◽

Operating Characteristic Curve

Abstract Background: This study aimed to analyze the incidence of early postpartum dyslipidemia and its potential predictors in women with a history of gestational diabetes mellitus (GDM).Methods: This was a retrospective study. 589 women diagnosed with GDM were enrolled and followed up at 6–12 weeks after delivery. A 75g oral glucose tolerance test (OGTT) and lipid levels were performed during mid-trimester and the early postpartum period. Participants were divided into the normal lipid group and dyslipidemia group according to postpartum lipid levels. Demographic and metabolic parameters were analyzed. Multiple logistic regression was performed to analyze the potential predictors for early postpartum dyslipidemia. A receiver operating characteristic curve (ROC) was calculated to determine the cut-off values.Results: A total of 38.5% of the 589 women developed dyslipidemia in early postpartum and 60% of them had normal glucose metabolism. Delivery age, systolic blood pressure (SBP), glycated hemoglobin (HbA1c) and low-density lipoprotein cholesterol (LDL-C) were independent predictors of early postpartum dyslipidemia in women with a history of GDM. The cut-offs of maternal age, SBP, HbA1c values, and LDL-C levels were 35 years, 123mmHg, 5.1%, and 3.56 mmol/L, respectively. LDL-C achieved a balanced mix of high sensitivity (63.9%) and specificity (69.2%), with the highest area under the receiver operating characteristic curve (AUC) (0.696). When LDL-C was combined with age, SBP, and HbA1c, the AUC reached to 0.733.Conclusions: A lipid metabolism evaluation should be recommended in women with a history of GDM after delivery, particularly those with a maternal age > 35 years, SBP > 123 mmHg before labor, HbA1c value > 5.1%, or LDL-C levels > 3.56 mmol/L in the second trimester of pregnancy.

Download Full-text

Intubated COVID-19 predictive (ICOP) score for early mortality after intubation in patients with COVID-19

Scientific Reports ◽

10.1038/s41598-021-00591-1 ◽

2021 ◽

Vol 11 (1) ◽

Author(s):

Mitsuaki Nishikimi ◽

Rehana Rasul ◽

Cristina P. Sison ◽

Daniel Jafari ◽

Muhammad Shoaib ◽

...

Keyword(s):

New York ◽

Operating Characteristic ◽

Characteristic Curve ◽

Early Mortality ◽

Predictive Score ◽

Derivation Cohort ◽

Risk Of Mortality ◽

Increased Risk ◽

Failure Assessment ◽

History Of

AbstractPatients with coronavirus disease 2019 (COVID-19) can have increased risk of mortality shortly after intubation. The aim of this study is to develop a model using predictors of early mortality after intubation from COVID-19. A retrospective study of 1945 intubated patients with COVID-19 admitted to 12 Northwell hospitals in the greater New York City area was performed. Logistic regression model using backward selection was applied. This study evaluated predictors of 14-day mortality after intubation for COVID-19 patients. The predictors of mortality within 14 days after intubation included older age, history of chronic kidney disease, lower mean arterial pressure or increased dose of required vasopressors, higher urea nitrogen level, higher ferritin, higher oxygen index, and abnormal pH levels. We developed and externally validated an intubated COVID-19 predictive score (ICOP). The area under the receiver operating characteristic curve was 0.75 (95% CI 0.73–0.78) in the derivation cohort and 0.71 (95% CI 0.67–0.75) in the validation cohort; both were significantly greater than corresponding values for sequential organ failure assessment (SOFA) or CURB-65 scores. The externally validated predictive score may help clinicians estimate early mortality risk after intubation and provide guidance for deciding the most effective patient therapies.

Download Full-text

Cardiopulmonary Exercise Testing in Patients with Chronic Heart Failure: Prognostic Comparison from Peak VO2 and VE/VCO2 Slope

The Open Cardiovascular Medicine Journal ◽

10.2174/1874192401004010127 ◽

2010 ◽

Vol 4 (1) ◽

pp. 127-134 ◽

Cited By ~ 36

Author(s):

Filippo Maria Sarullo ◽

Giovanni Fazio ◽

Ignazio Brusca ◽

Sergio Fasullo ◽

Salvatore Paterna ◽

...

Keyword(s):

Heart Failure ◽

Chronic Heart Failure ◽

Risk Stratification ◽

Operating Characteristic ◽

Characteristic Curve ◽

Cardiopulmonary Exercise Testing ◽

Cardiopulmonary Exercise ◽

Peak Vo2 ◽

Operating Characteristic Curve ◽

Related Mortality

Background: Cardiopulmonary exercise testing with ventilatory expired gas analysis (CPET) has proven to be a valuable tool for assessing patients with chronic heart failure (CHF). The maximal oxygen uptake (peak V02) is used in risk stratification of patients with CHF. The minute ventilation-carbon dioxide production relationship (VE/VCO2 slope) has recently demonstrated prognostic significance in patients with CHF. Methods: Between January 2006 and December 2007 we performed CPET in 184 pts (146 M, 38 F, mean age 59.8 ± 12.9 years), with stable CHF (96 coronary artery disease, 88 dilated cardiomyopathy), in NYHA functional class II (n.107) - III (n.77), with left ventricular ejection fraction (LVEF) ≤ 45%,. The ability of peak VO2 and VE/VCO2 slope to predict cardiac related mortality and cardiac related hospitalization within 12 months after evaluation was examined. Results: Peak VO2 and VE/VCO2 slope were demonstrated with univariate Cox regression analysis both to be significant predictor of cardiac-related mortality and hospitalization (p < 0.0001, respectively). Non survivors had a lower peak VO2 (10.49 ± 1.70 ml/kg/min vs. 14.41 ± 3.02 ml/kg/min, p < 0.0001), and steeper Ve/VCO2 slope (41.80 ± 8.07 vs. 29.84 ± 6.47, p < 0.0001) than survivors. Multivariate survival analysis revealed that VE/VCO2 slope added additional value to VO2 peak as an independent prognostic factor (χ2: 56.48, relative risk: 1.08, 95% CI: 1.03 – 1.13, p = 0.001). The results from Kaplan-Meier analysis revealed a 1-year cardiac-related mortality of 75% in patients with VE/VCO2 slope ≥ 35.6 and 25% in those with VE/VCO2 slope < 35.6 (log rank χ2: 67.03, p < 0.0001) and 66% in patients with peak VO2 ≤ 12.2 ml/kg/min and 34% in those with peak VO2 > 12.2 ml/kg/min (log rank χ2: 50.98, p < 0.0001). One-year cardiac-related hospitalization was 77% in patients with VE/VCO2 slope ≥ 32.5 and 23% in those with VE/VCO2 slope < 32.5 (log rank χ2: 133.80, p < 0.0001) and 63% in patients with peak VO2 ≤ 12.3 ml/kg/min and 37% in those with peak VO2 > 12.3 ml/kg/min (log rank χ2: 72.86, p < 0.0001). The VE/VCO2 slope was demonstrated with receiver operating characteristic curve analysis to be equivalent to peak VO2 in predicting cardiac-related mortality (0.89 vs. 0.89). Although area under the receiver operating characteristic curve for the VE/VCO2 slope was greater than peak VO2 in predicting cardiac-related hospitalization (0.88 vs 0.82), the difference was no statistically significant (p = 0.13). Conclusion: These results add to the present body of knowledge supporting the use of CPET in CHF patients. The VE/VCO2 slope, as an index of ventilatory response to exercise, is an excellent prognostic parameter and improves the risk stratification of CHF patients. It is easier to obtain than parameters of maximal exercise capacity and is of equivalent prognostic importance than peak VO2.

Download Full-text

Predictors associated with increased troponin in acute decompensated and chronic heart failure patients

Revista Romana de Medicina de Laborator ◽

10.2478/rrlm-2021-0015 ◽

2021 ◽

Vol 29 (2) ◽

pp. 153-164

Author(s):

Anamaria Draghici ◽

Catalin Adrian Buzea ◽

Caterina Delcea ◽

Ancuta Vijan ◽

Gheorghe Andrei Dan

Keyword(s):

Heart Failure ◽

Myocardial Injury ◽

Tertiary Care ◽

Multivariable Analysis ◽

High Sensitivity ◽

Care Hospital ◽

History Of ◽

Coronary Syndromes ◽

Kidney Disease Progression ◽

Male Sex

Abstract Background: Myocardial injury (INJ) expressed by elevated high-sensitivity troponin (hs-Tn) is common in heart failure (HF), due to cardiovascular and non-cardiac conditions. The mechanisms of INJ in acute decompensated HF (ADHF) versus chronic HF (CHF) are still debated. This study’s purpose was to evaluate the determinants of elevated hs-TnT in ADHF and CHF. Methods: We retrospectively analyzed consecutive HF patients with hs-TnT measured on admission, hospitalized in a tertiary-care hospital. Rehospitalizations, acute coronary syndromes, embolisms, infections, autoimmunity and malignancy were excluded. Cut-off point for hs-TnT was 14 ng/L. Results: Our study included 488 HF patients, 56.55% with ADHF. Mean age was 72.52±10.09 years. 53.89% were females. 67.75% ADHF and 45.75% CHF patients had elevated hs-TnT. Median hs-TnT was higher in ADHF versus CHF (21.05[IQR 12.74-33.81] vs 13.20[IQR 7.93-23.25], p<0.0001). In multivariable analysis in ADHF and CHF, log10NT-proBNP (HR=5.30, 95%CI 2.71–10.38, p<0.001, respectively HR=5.49, 95%CI 1.71–17.57, p=0.004) and eGFR (HR=0.72, 95%CI 0.62–0.85, p<0.001, respectively HR=0.71, 95%CI 0.55–0.93, p=0.014) were independent predictors for increased hs-TnT. Independent factors associated with elevated hs-TnT in ADHF were male sex (HR=2.52, 95%CI 1.31-4.87, p=0.006) and chronic pulmonary obstructive disease (COPD) (HR=10.57, 95%CI 1.26-88.40, p=0.029), while in CHF were age (HR=2.68, 95%CI 1.42-5.07, p=0.002) and previous stroke (HR=5.35, 95%CI 0.98-29.20, p=0.053). Conclusion: HF severity, expressed by NT-proBNP levels, and kidney disease progression, expressed by eGFR, were independent predictors associated with increased hs-TnT in both ADHF and CHF. Specific independent predictors were also indentified in ADHF (male sex, COPD) and CHF (age, history of stroke).

Download Full-text

Risk Factors and Predictive Value of Perioperative Neurocognitive Disorders in Elderly Patients With Gastrointestinal Tumors

10.21203/rs.3.rs-244830/v1 ◽

2021 ◽

Author(s):

Yongli Li ◽

Hui-Fan Huang ◽

Yuan Le

Keyword(s):

Risk Factors ◽

Elderly Patients ◽

Predictive Value ◽

Operating Characteristic ◽

Characteristic Curve ◽

Alcohol Exposure ◽

Curve Analysis ◽

Vas Score ◽

History Of ◽

Independent Risk Factors

Abstract Background: This study aims to investigate the risk factors of perioperative neurocognitive disorders (PNDs) mainly including postoperative cognitive dysfunction (POCD) in elderly patients with gastrointestinal tumors, and evaluate its predictive value.Methods: A total of 222 eligible elderly patients (≥65 years) scheduled for elective gastroenterectomy under general anesthesia were enrolled. The cognitive function assessment was carried out 1 day before surgery and 7 days after surgery. Receiver operating characteristic curve analysis was performed to evaluate the predictive value of risk factors for early POCD. The risk factors for POCD were analyzed using a multivariate logistic regression mode.Results: Of all the 222 enrolled patients, 91 (41.0%) developed early POCD and 40 (18.0%) were identified as major POCD within 7 days after the surgery. Visual analogue score (VAS, 1st day, resting) ≥4 (OR=7.618[3.231–17.962], P<0.001) and alcohol exposure (OR= 2.398[1.174–4.900], P=0.016) were independent risk factors for early POCD. VAS score (1st, resting) ≥4 (OR=13.823[4.779–39.981], P<0.001), preoperative white blood cell (WBC) levels ≥10x10*9/L(OR=5.548[1.128-26.221], P=0.035), blood loss ≥500ml (OR=3.317[1.094-10.059], P=0.034), history of hypertension (OR=3.046[1.267-7.322], P=0.013), and neutrophil–lymphocyte ratio (NLR) ≥2 (OR=3.261[1.020-10.419], P=0.046) were independent risk factors for major POCD. Receiver operating characteristic curve analysis indicated that VAS score (1st day, resting) was a significant predictor for major POCD with a cut-off value of 2.68 and an area under the curve of 0.860 (95% confidence interval: 0.801–0.920, P<0.001).Conclusions: The risk factors for early POCD after gastroenterectomy included high VAS score (1st day, resting) and alcohol exposure. High VAS score, preoperative WBC levels ≥10x10*9/L, blood loss ≥500ml, NLR ≥2, and history of hypertension were independent risk factors for major POCD. Among them, VAS score had a high predictive value.

Download Full-text

Early postpartum dyslipidemia and its potential predictors during pregnancy in women with a history of gestational diabetes mellitus

Lipids in Health and Disease ◽

10.1186/s12944-020-01398-1 ◽

2020 ◽

Vol 19 (1) ◽

Author(s):

Ling Pei ◽

Huangmeng Xiao ◽

Fenghua Lai ◽

Zeting Li ◽

Zhuyu Li ◽

...

Keyword(s):

Diabetes Mellitus ◽

Gestational Diabetes ◽

Gestational Diabetes Mellitus ◽

Maternal Age ◽

Operating Characteristic ◽

Characteristic Curve ◽

Lipid Levels ◽

Early Postpartum ◽

History Of ◽

Operating Characteristic Curve

Abstract Background This study aimed to analyze the incidence of early postpartum dyslipidemia and its potential predictors in women with a history of gestational diabetes mellitus (GDM). Methods This was a retrospective study. Five hundred eighty-nine women diagnosed with GDM were enrolled and followed up at 6–12 weeks after delivery. A 75 g oral glucose tolerance test (OGTT) and lipid levels were performed during mid-trimester and the early postpartum period. Participants were divided into the normal lipid group and dyslipidemia group according to postpartum lipid levels. Demographic and metabolic parameters were analyzed. Multiple logistic regression was performed to analyze the potential predictors for early postpartum dyslipidemia. A receiver operating characteristic curve (ROC) was calculated to determine the cut-off values. Results A total of 38.5% of the 589 women developed dyslipidemia in early postpartum and 60% of them had normal glucose metabolism. Delivery age, systolic blood pressure (SBP), glycated hemoglobin (HbA1c) and low-density lipoprotein cholesterol (LDL-C) were independent predictors of early postpartum dyslipidemia in women with a history of GDM. The cut-offs of maternal age, SBP, HbA1c values, and LDL-C levels were 35 years, 123 mmHg, 5.1%, and 3.56 mmol/L, respectively. LDL-C achieved a balanced mix of high sensitivity (63.9%) and specificity (69.2%), with the highest area under the receiver operating characteristic curve (AUC) (0.696). When LDL-C was combined with age, SBP, and HbA1c, the AUC reached to 0.733. Conclusions A lipid metabolism evaluation should be recommended in women with a history of GDM after delivery, particularly those with a maternal age > 35 years, SBP > 123 mmHg before labor, HbA1c value > 5.1%, or LDL-C levels > 3.56 mmol/L in the second trimester of pregnancy.

Download Full-text

Conventional administration and scoring procedures suppress the diagnostic accuracy of a performance-based test designed to assess balance ability in lower limb prosthesis users

Prosthetics and Orthotics International ◽

10.1177/0309364619848274 ◽

2019 ◽

Vol 43 (4) ◽

pp. 402-408 ◽

Cited By ~ 2

Author(s):

Andrew Sawers ◽

Brian J Hafner

Keyword(s):

Diagnostic Accuracy ◽

Receiver Operating Characteristic Curve ◽

Receiver Operating Characteristic ◽

Operating Characteristic ◽

Characteristic Curve ◽

Practice Effects ◽

Balance Test ◽

Walking Test ◽

History Of ◽

Operating Characteristic Curve

Background:Practice effects have been observed among performance-based clinical tests administered to prosthesis-users. Their impact on test applications remains unknown.Objective:To determine whether scoring a clinical balance test using conventional procedures that do not accommodate practice effects reduces its diagnostic accuracy relative to scoring it using recommended procedures that do accommodate practice effects.Study Design:Cross-sectional study.Methods:Narrowing Beam Walking Test data from 40 prosthesis users was scored using recommended methods (i.e. average of trials 3–5) and conventional methods applied to other tests (i.e. mean or best of trials 1–3). Area under the receiver operating characteristic curve for each method was compared to 0.50, to determine if it was better than chance at identifying prosthesis-users with a history of falls, and to 0.80, to determine if it surpassed a threshold recommended for diagnostic accuracy.Results:Receiver operating characteristic curve area decreased when the Narrowing Beam Walking Test was scored using conventional rather than recommended procedures. Furthermore, when scored using conventional procedures, the NBWT no longer discriminated between prosthesis-users with and without a history of falls with a probability greater than chance, or exceeded recommended diagnostic thresholds.Conclusion:Scoring the Narrowing Beam Walking Test using conventional procedures that do not accommodate practice effects decreased its diagnostic accuracy among prosthesis-users relative to recommended procedures. Conventional scoring procedures may limit the effectiveness of performance-based tests used to screen for fall risk in prosthesis-users because they do not mitigate practice effects. The influence of practice effects on other tests, and test applications (e.g. clinical evaluation and prediction), is warranted.Clinical relevanceScoring a clinical balance test using conventional procedures that do not mitigate practice effects reduced its diagnostic accuracy. Changing administration and scoring procedures to accommodate practice effects should be considered to improve the diagnostic accuracy of other performance-based balance tests.

Download Full-text

Potential of deep learning in assessing pneumoconiosis depicted on digital chest radiography

Occupational and Environmental Medicine ◽

10.1136/oemed-2019-106386 ◽

2020 ◽

Vol 77 (9) ◽

pp. 597-602

Author(s):

Xiaohua Wang ◽

Juezhao Yu ◽

Qiao Zhu ◽

Shuqiang Li ◽

Zanmei Zhao ◽

...

Keyword(s):

Deep Learning ◽

Operating Characteristic ◽

Characteristic Curve ◽

Classification Performance ◽

X Ray ◽

Testing Dataset ◽

Learning Techniques ◽

History Of ◽

Digital Chest ◽

Chest X Ray

ObjectivesTo investigate the potential of deep learning in assessing pneumoconiosis depicted on digital chest radiographs and to compare its performance with certified radiologists.MethodsWe retrospectively collected a dataset consisting of 1881 chest X-ray images in the form of digital radiography. These images were acquired in a screening setting on subjects who had a history of working in an environment that exposed them to harmful dust. Among these subjects, 923 were diagnosed with pneumoconiosis, and 958 were normal. To identify the subjects with pneumoconiosis, we applied a classical deep convolutional neural network (CNN) called Inception-V3 to these image sets and validated the classification performance of the trained models using the area under the receiver operating characteristic curve (AUC). In addition, we asked two certified radiologists to independently interpret the images in the testing dataset and compared their performance with the computerised scheme.ResultsThe Inception-V3 CNN architecture, which was trained on the combination of the three image sets, achieved an AUC of 0.878 (95% CI 0.811 to 0.946). The performance of the two radiologists in terms of AUC was 0.668 (95% CI 0.555 to 0.782) and 0.772 (95% CI 0.677 to 0.866), respectively. The agreement between the two readers was moderate (kappa: 0.423, p<0.001).ConclusionOur experimental results demonstrated that the deep leaning solution could achieve a relatively better performance in classification as compared with other models and the certified radiologists, suggesting the feasibility of deep learning techniques in screening pneumoconiosis.

Download Full-text