A deep learning model for predicting COVID-19 ARDS in critically ill patients

Predictive Modeling of Acute Respiratory Distress Syndrome Using Machine Learning: Systematic Review and Meta-Analysis

BACKGROUND

Acute respiratory distress syndrome (ARDS) is a critical condition commonly encountered in the intensive care unit (ICU), characterized by a high incidence and substantial mortality rate. Early detection and accurate prediction of ARDS can significantly improve patient outcomes. While machine learning (ML) models are increasingly being used for ARDS prediction, there is a lack of consensus on the most effective model or methodology. This study is the first to systematically evaluate the performance of ARDS prediction models based on multiple quantitative data sources. We compare the effectiveness of ML models via a meta-analysis, revealing factors affecting performance and suggesting strategies to enhance generalization and prediction accuracy.

OBJECTIVE

This study aims to evaluate the performance of existing ARDS prediction models through a systematic review and meta-analysis, using metrics such as area under the receiver operating characteristic curve, sensitivity, specificity, and other relevant indicators. The findings will provide evidence-based insights to support the development of more accurate and effective ARDS prediction tools.

METHODS

We performed a search across 6 electronic databases for studies developing ML predictive models for ARDS, with a cutoff date of December 29, 2024. The risk of bias in these models was evaluated using the Prediction model Risk of Bias Assessment Tool. Meta-analyses and investigations into heterogeneity were carried out using Meta-DiSc software (version 1.4), developed by the Ramón y Cajal Hospital's Clinical Biostatistics team in Madrid, Spain. Furthermore, sensitivity, subgroup, and meta-regression analyses were used to explore the sources of heterogeneity more comprehensively.

RESULTS

ML models achieved a pooled area under the receiver operating characteristic curve of 0.7407 for ARDS. The additional metrics were as follows: sensitivity was 0.67 (95% CI 0.66-0.67; P<.001; I²=97.1%), specificity was 0.68 (95% CI 0.67-0.68; P<.001; I²=98.5%), the diagnostic odds ratio was 6.26 (95% CI 4.93-7.94; P<.001; I²=95.3%), the positive likelihood ratio was 2.80 (95% CI 2.46-3.19; P<.001; I²=97.3%), and the negative likelihood ratio was 0.51 (95% CI 0.46-0.57; P<.001; I²=93.6%).

CONCLUSIONS

This study evaluates prediction models constructed using various ML algorithms, with results showing that ML demonstrates high performance in ARDS prediction. However, many of the existing models still have limitations. During model development, it is essential to focus on model quality, including reducing bias risk, designing appropriate sample sizes, conducting external validation, and ensuring model interpretability. Additionally, challenges such as physician trust and the need for prospective validation must also be addressed. Future research should standardize model development, optimize model performance, and explore how to better integrate predictive models into clinical practice to improve ARDS diagnosis and risk stratification.

TRIAL REGISTRATION

PROSPERO CRD42024529403; https://www.crd.york.ac.uk/PROSPERO/view/CRD42024529403.

Accuracy of artificial intelligence algorithms in predicting acute respiratory distress syndrome: a systematic review and meta-analysis

BACKGROUND

Acute respiratory distress syndrome (ARDS) is a serious threat to human life. Hence, early and accurate diagnosis and treatment are crucial for patient survival. This meta-analysis evaluates the accuracy of artificial intelligence in the early diagnosis of ARDS and provides guidance for future research and applications.

METHODS

A search on PubMed, Embase, Cochrane, Web of Science, CNKI, Wanfang, Chinese Biomedical Literature (CBM), and VIP databases was systematically conducted, from their establishment to November 2023, to obtain eligible studies for the analysis and evaluation of the predictive effect of AI on ARDS. The retrieved literature was screened according to inclusion and exclusion criteria, the quality of the included studies was assessed using QUADAS-2, and the included studies were statistically analyzed.

RESULTS

Among the 2, 996 studies, 33 were included in this meta-analysis, which showed that the pooled sensitivity of AI in predicting ARDS was 0.81 (0.76-0.85), the pooled specificity was 0.88 (0.84-0.91), and the area under the receiver operating characteristic curve (AUC) was 0.91 (0.88-0.93). The analyzed studies included 28 models, with a pooled sensitivity of 0.79 (0.76-0.82), a pooled specificity of 0.85 (0.83-0.88), and an AUC of 0.89 (0.86-0.91). In the subgroup analysis, the pooled AUC of the AI models ANN, CNN, LR, RF, SVM, and XGB were 0.86 (0.83-0.89), 0.91 (0.88-0.93), 0.86 (0.83-0.89), and 0.89 (0.86-0.91), 0.90 (0.87-0.92), 0.93 (0.90-0.95), respectively. In an additional subgroup analysis, we evaluated the predictive performance of the AI models trained using different predictors. This meta-analysis was registered in PROSPERO (CRD42023491546).

CONCLUSION

AI has good sensitivity and specificity for predicting ARDS, indicating a high clinical application value. Algorithmic models such as CNN, SVM, and XGB have improved prediction performance. The subgroup analysis revealed that the model trained using images combined with other predictors had the best predictive performance.

Using the Regression Slope of Training Loss to Optimize Chest X-ray Generation in Deep Convolutional Generative Adversarial Networks

Diffusion models, variational autoencoders, and generative adversarial networks (GANs) are three common types of generative artificial intelligence models for image generation. Among these, GANs are the most frequently used for medical image generation and are often employed for data augmentation in various studies. However, due to the adversarial nature of GANs, where the generator and discriminator compete against each other, the training process can sometimes end with the model unable to generate meaningful images or even producing noise. This phenomenon is rarely discussed in the literature, and no studies have proposed solutions to address this issue. Such outcomes can introduce significant bias when GANs are used for data augmentation in medical image training. Moreover, GANs often require substantial computational power and storage, adding to the challenges. In this study, we used deep convolutional GANs for chest X-ray generation, and three typical training outcomes were found. Two scenarios generated meaningful medical images and one failed to produce usable images. By analyzing the loss history during training, we observed that the regression line of the overall losses tends to diverge slowly. After excluding outlier losses, we found that the slope of the regression line within the stable loss segment indicates the optimal point to terminate training, ensuring the generation of meaningful medical images.

Development and validation of a deep learning-based framework for automated lung CT segmentation and acute respiratory distress syndrome prediction: a multicenter cohort study

Background

Acute respiratory distress syndrome (ARDS) is a life-threatening condition with a high incidence and mortality rate in intensive care unit (ICU) admissions. Early identification of patients at high risk for developing ARDS is crucial for timely intervention and improved clinical outcomes. However, the complex pathophysiology of ARDS makes early prediction challenging. This study aimed to develop an artificial intelligence (AI) model for automated lung lesion segmentation and early prediction of ARDS to facilitate timely intervention in the intensive care unit.

Methods

A total of 928 ICU patients with chest computed tomography (CT) scans were included from November 2018 to November 2021 at three centers in China. Patients were divided into a retrospective cohort for model development and internal validation, and three independent cohorts for external validation. A deep learning-based framework using the UNet Transformer (UNETR) model was developed to perform the segmentation of lung lesions and early prediction of ARDS. We employed various data augmentation techniques using the Medical Open Network for AI (MONAI) framework, enhancing the training sample diversity and improving the model's generalization capabilities. The performance of the deep learning-based framework was compared with a Densenet-based image classification network and evaluated in external and prospective validation cohorts. The segmentation performance was assessed using the Dice coefficient (DC), and the prediction performance was assessed using area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. The contributions of different features to ARDS prediction were visualized using Shapley Explanation Plots. This study was registered with the China Clinical Trial Registration Centre (ChiCTR2200058700).

Findings

The segmentation task using the deep learning framework achieved a DC of 0.734 ± 0.137 in the validation set. For the prediction task, the deep learning-based framework achieved AUCs of 0.916 [0.858-0.961], 0.865 [0.774-0.945], 0.901 [0.835-0.955], and 0.876 [0.804-0.936] in the internal validation cohort, external validation cohort I, external validation cohort II, and prospective validation cohort, respectively. It outperformed the Densenet-based image classification network in terms of prediction accuracy. Moreover, the ARDS prediction model identified lung lesion features and clinical parameters such as C-reactive protein, albumin, bilirubin, platelet count, and age as significant contributors to ARDS prediction.

Interpretation

The deep learning-based framework using the UNETR model demonstrated high accuracy and robustness in lung lesion segmentation and early ARDS prediction, and had good generalization ability and clinical applicability.

Funding

This study was supported by grants from the Shanghai Renji Hospital Clinical Research Innovation and Cultivation Fund (RJPY-DZX-008) and Shanghai Science and Technology Development Funds (22YF1423300).