Lung image segmentation based on DRD U-Net and combined WGAN with Deep Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• COVID-19 poses a significant global health threat, necessitating improved diagnostic methods.
• Current COVID-19 detection methods have limitations that deep learning can address.
• Accurate image-based diagnosis is crucial for timely patient management.

Purpose of the Study:

• To propose a multi-classification deep learning model for COVID-19 diagnosis.
• To enhance diagnostic capabilities by integrating segmentation and classification tasks.
• To improve the differentiation between COVID-19, other pneumonias, and healthy individuals using medical images.

Main Methods:

• Employed an end-to-end DRD U-Net model for segmenting lung lesions, optimizing feature reuse.
• Utilized a combination of Generative Adversarial Networks (WGAN) and Deep Neural Network (DNN) classifiers for multi-class image classification.
• Addressed challenges of small sample sizes in COVID-19 datasets for effective classification.

Main Results:

• The segmentation task demonstrated optimal performance in Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD) metrics.
• Classification accuracy improved from 65.32% to 73.84% compared to SMOTE oversampling.
• Achieved superior results in F-measure (74.65%) and G-mean (74.37%) and showed advantages over other multi-task models.

Conclusions:

• The proposed multi-task deep learning model offers a promising approach for COVID-19 image diagnosis.
• The study enhances diagnostic accuracy and provides a potential tool for clinical application.
• This method could significantly aid in the identification and management of COVID-19 patients.