Unsupervised Deep Learning based Variational Autoencoder Model for COVID-19 Diagnosis and Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• COVID-19 poses a global health challenge, with rapid spread complicating control efforts.
• Limited availability of testing kits hinders effective disease management and diagnosis.
• Artificial intelligence (AI), including machine learning (ML) and deep learning (DL), shows potential for COVID-19 analysis and prediction.

Purpose of the Study:

• To introduce a novel unsupervised deep learning variational autoencoder (UDL-VAE) model for COVID-19 detection and classification.
• To enhance image quality for improved diagnostic accuracy using AI techniques.
• To leverage unsupervised learning within DL for proficient COVID-19 prediction.

Main Methods:

• Implemented an unsupervised deep learning variational autoencoder (UDL-VAE) model.
• Utilized adaptive Wiener filtering (AWF) for image preprocessing and quality enhancement.
• Employed Inception v4 with Adagrad as a feature extractor for the classification task.

Main Results:

• The UDL-VAE model demonstrated superior diagnostic performance in experimental evaluations.
• Achieved high accuracy rates of 0.987 for binary classification and 0.992 for multiple-class classification.
• The model effectively detects and classifies COVID-19 from medical images.

Conclusions:

• The proposed UDL-VAE model offers an effective and accurate method for COVID-19 detection and classification.
• Unsupervised deep learning techniques can be successfully integrated into DL models for improved disease prediction.
• This AI-based approach provides a valuable tool to aid in managing the global COVID-19 pandemic.