Variational Autoencoder Based Imbalanced COVID-19 Detection Using Chest X-Ray Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Early COVID-19 detection is critical for pandemic management.
• Deep learning on chest X-rays shows promise but struggles with imbalanced datasets.
• Existing models often fail to adequately address class imbalance issues.

Purpose of the Study:

• To develop a robust method for detecting COVID-19, Pneumonia, and Normal cases from chest X-rays.
• To overcome the challenge of imbalanced datasets in medical image classification.
• To improve the accuracy and reliability of deep learning models for disease detection.

Main Methods:

• Utilized unsupervised Variational Auto Encoders (VAEs) to learn salient features from chest X-ray images and map them to a latent space.
• Applied various data resampling techniques to balance imbalanced classes within the latent vector representation.
• Trained established classification models on the balanced, latent-space dataset for multi-class classification.
• Employed a 10-fold cross-validation technique to evaluate the effectiveness of resampling methods.

Main Results:

• The proposed VAE-based method demonstrated significant improvements in COVID-19 detection accuracy.
• Balancing the dataset in the latent space enhanced the performance of classification models.
• Wilcoxon rank test confirmed the statistical significance of the obtained results at a 95% confidence level.

Conclusions:

• Variational Auto Encoders offer an effective solution for handling imbalanced datasets in chest X-ray-based disease detection.
• The VAE-enhanced approach improves the classification of COVID-19, Pneumonia, and Normal cases.
• This method provides a promising direction for more accurate and reliable early disease detection systems.