Multitask Adversarial Networks Based on Extensive Nonlinear Spiking Neuron Models.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Deep learning models show promise in analyzing COVID-19 Chest X-ray (CXR) images.
• Challenges include low image quality, limited data, complex features, and irregular shapes in COVID-19 pneumonia.
• Existing deep learning methods struggle with these inherent limitations.

Purpose of the Study:

• To develop an advanced deep learning architecture for improved COVID-19 CXR analysis.
• To address challenges of low image quality and classification accuracy in COVID-19 detection.
• To introduce a multitask adversarial network (MAE-Net) for enhanced CXR image processing and classification.

Main Methods:

• Proposed a novel multitask adversarial network (MAE-Net) utilizing an extensive NSNP-like neuron model.
• MAE-Net performs dual tasks: enhancing low-quality CXR images and classifying COVID-19 cases.
• Employed an adversarial architecture with two generators, two discriminators, and two novel loss functions.

Main Results:

• MAE-Net demonstrated superior performance in enhancing CXR image quality.
• The model achieved higher accuracy in classifying COVID-19 cases compared to eight other deep learning models.
• Experiments were conducted on four benchmark COVID-19 CXR datasets.

Conclusions:

• The proposed MAE-Net effectively overcomes limitations in COVID-19 CXR analysis.
• The model significantly improves both image conversion quality and classification accuracy.
• MAE-Net offers a promising solution for AI-driven COVID-19 diagnosis using CXR images.