Combining chest X-rays and electronic health record (EHR) data using machine learning to diagnose acute respiratory failure

Jabbour, Sarah; Fouhey, David F.; Kazerooni, Ella A.; Wiens, Jenna; Sjoding, Michael W.

doi:10.1093/jamia/ocac030

Cited by 26 publications

(22 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our results are in line with a similar study by Jabbour et al, who developed machine learning models combining CXRs and clinical data to identify acute respiratory failure. 39 The results presented in our combined model with clinical data and images had better sensitivity, specificity, and AUC than the separate clinical data model and image models. We found the evidence that seems to suggest the benefit of combined machine learning models and could be comparable with a clinician's judgment.…”

Section: Discussionmentioning

confidence: 74%

Artificial intelligence–aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs

et al. 2022

View full text Add to dashboard Cite

Objective The aim of this study was to develop an artificial intelligence–based model to detect the presence of acute respiratory distress syndrome (ARDS) using clinical data and chest X-ray (CXR) data. Method The transfer learning method was used to train a convolutional neural network (CNN) model with an external image dataset to extract the image features. Then, the last layer of the model was fine-tuned to determine the probability of ARDS. The clinical data were trained using three machine learning algorithms—eXtreme Gradient Boosting (XGB), random forest (RF), and logistic regression (LR)—to estimate the probability of ARDS. Finally, ensemble-weighted methods were proposed that combined the image model and the clinical data model to estimate the probability of ARDS. An analysis of the importance of clinical features was performed to explore the most important features in detecting ARDS. A gradient-weighted class activation mapping (Grad-CAM) model was used to explain what our CNN sees and understands when making a decision. Results The proposed ensemble-weighted methods improved the performances of the ARDS classifiers (XGB + CNN, area under the curve [AUC] = 0.916; RF + CNN, AUC = 0.920; LR + CNN, AUC = 0.920; XGB + RF + LR + CNN, AUC = 0.925). In addition, the ML model using clinical data to present the top 15 important features to identify the risk factors of ARDS. Conclusion This study developed combined machine learning models with clinical data and CXR images to detect ARDS. According to the results of the Shapley Additive exPlanations values and the Grad-CAM techniques, an explicable ARDS diagnosis model is suitable for a real-life scenario.

show abstract

Section: Discussionmentioning

confidence: 74%

Artificial intelligence–aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Examples of observed 'spurious correlation' includes, sex changes, the appearance of various support devices, as well as changing body position (relating to switches from AP to PA projection). 14 Highlighting the influence of confounders and biases in predictions made by learning-based systems is essential for building safer and fairer predictive models. This is especially relevant for translating learning-based computer-aided diagnostic/screening systems to routine clinical care.…”

Section: Discussionmentioning

confidence: 99%

Learning disentangled representations for explainable chest x-ray classification using Dirichlet VAEs

Harkness

Frangi

Zucker

et al. 2023

Medical Imaging 2023: Image Processing

View full text Add to dashboard Cite

This study explores the use of the Dirichlet Variational Autoencoder (DirVAE) for learning disentangled latent representations of chest X-ray (CXR) images. Our working hypothesis is that distributional sparsity, as facilitated by the Dirichlet prior, will encourage disentangled feature learning for the complex task of multi-label classification of CXR images. The DirVAE is trained using CXR images from the CheXpert database, and the predictive capacity of multi-modal latent representations learned by DirVAE models is investigated through implementation of an auxiliary multi-label classification task, with a view to enforce separation of latent factors according to class-specific features. The predictive performance and explainability of the latent space learned using the DirVAE were quantitatively and qualitatively assessed, respectively, and compared with a standard Gaussian prior-VAE (GVAE). We introduce a new approach for explainable multi-label classification in which we conduct gradient-guided latent traversals for each class of interest. Study findings indicate that the DirVAE is able to disentangle latent factors into class-specific visual features, a property not afforded by the GVAE, and achieve a marginal increase in predictive performance relative to GVAE. We generate visual examples to show that our explainability method, when applied to the trained DirVAE, is able to highlight regions in CXR images that are clinically relevant to the class(es) of interest and additionally, can identify cases where classification relies on spurious feature correlations.

show abstract

“… 38 , 39 , 40 , 41 , 42 , 43 , 44 The likelihood of such a bias is particularly high when the data per class originates from different sources, such as different countries, hospitals, or imaging systems. 45 , 46 In these cases, underlying differences in the image data distributions, due to for example a difference in image acquisition parameters, post‐processing operations, or overall patient characteristics unrelated to COVID‐19, might create spurious correlations. Especially when these differences are more obvious than the COVID‐19 disease features, they are likely to be exploited by the neural network (NN).…”

Section: Introductionmentioning

confidence: 99%

“…However, this approach can increase the risk of hidden biases that may lead to overly optimistic results 38–44 . The likelihood of such a bias is particularly high when the data per class originates from different sources, such as different countries, hospitals, or imaging systems 45,46 . In these cases, underlying differences in the image data distributions, due to for example a difference in image acquisition parameters, post‐processing operations, or overall patient characteristics unrelated to COVID‐19, might create spurious correlations.…”

Section: Introductionmentioning

confidence: 99%

“…47,48 Jabbour et al for example showed that NNs can accurately identify patient attributes in CXRs such as sex and age, and the NNs tend to exploit correlations between these attributes and the outcome label when learning to predict a diagnosis, leading to poor performance when such correlations do not hold in the test population. 46 In a study by Kim et al, underlying differences in dataset distributions of commonly used COVID-19 CXR datasets were visualized through principal component analysis and t-distributed stochastic neighbor embedding. 48 Excellent performance during internal validation and poor performance in external validation showed these differences were likely exploited by the NNs during training.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Automatic coronavirus disease 2019 diagnosis based on chest radiography and deep learning – Success story or dataset bias?

2022

View full text Add to dashboard Cite

Purpose: Over the last 2 years, the artificial intelligence (AI) community has presented several automatic screening tools for coronavirus disease 2019 based on chest radiography (CXR), with reported accuracies often well over 90%. However, it has been noted that many of these studies have likely suffered from dataset bias, leading to overly optimistic results. The purpose of this study was to thoroughly investigate to what extent biases have influenced the performance of a range of previously proposed and promising convolutional neural networks (CNNs), and to determine what performance can be expected with current CNNs on a realistic and unbiased dataset. Methods: Five CNNs for COVID-19 positive/negative classification were implemented for evaluation, namely VGG19, ResNet50, InceptionV3, DenseNet201, and COVID-Net. To perform both internal and cross-dataset evaluations, four datasets were created. The first dataset Valencian Region Medical Image Bank (BIMCV) followed strict reverse transcriptase-polymerase chain reaction (RT-PCR) test criteria and was created from a single reliable open access databank, while the second dataset (COVIDxB8) was created through a combination of six online CXR repositories. The third and fourth datasets were created by combining the opposing classes from the BIMCV and COVIDxB8 datasets. To decrease inter-dataset variability, a pre-processing workflow of resizing, normalization, and histogram equalization were applied to all datasets. Classification performance was evaluated on unseen test sets using precision and recall. A qualitative sanity check was performed by evaluating saliency maps displaying the top 5%, 10%, and 20% most salient segments in the input CXRs, to evaluate whether the CNNs were using relevant information for decision making. In an additional experiment and to further investigate the origin of potential dataset bias, all pixel values outside the lungs were set to zero through automatic lung segmentation before training and testing. Results: When trained and evaluated on the single online source dataset (BIMCV), the performance of all CNNs is relatively low (precision: 0.65-0.72, recall: 0.59-0.71), but remains relatively consistent during external evaluation (precision: 0.58-0.82, recall: 0.57-0.72). On the contrary, when trained and internally evaluated on the combinatory datasets, all CNNs performed well across all metrics (precision: 0.94-1.00, recall: 0.77-1.00). However, when subsequently evaluated cross-dataset, results dropped substantially (precision: 0.10-0.61, recall: 0.04-0.80). For all datasets, saliency maps revealed the CNNs rarelyThis is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

show abstract

Combining chest X-rays and electronic health record (EHR) data using machine learning to diagnose acute respiratory failure

Cited by 26 publications

References 30 publications

Artificial intelligence–aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs

Artificial intelligence–aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs

Learning disentangled representations for explainable chest x-ray classification using Dirichlet VAEs

Automatic coronavirus disease 2019 diagnosis based on chest radiography and deep learning – Success story or dataset bias?

Contact Info

Product

Resources

About