MixNet-LD: An Automated Classification System for Multiple Lung Diseases Using Modified MixNet Model

Area of Science:

• Medical Imaging and Artificial Intelligence
• Computer-Aided Diagnosis
• Respiratory Medicine

Background:

• Lung diseases pose significant health risks, necessitating accurate severity assessment for effective management.
• Current treatment focuses on controlling irreversible lung disease severity.
• Automated, consistent methods are needed for reliable lung illness intensity determination.

Purpose of the Study:

• To develop an automated approach, MixNet-LD, for identifying and categorizing lung disease severity.
• To leverage an upgraded pre-trained MixNet model for enhanced lung illness classification.
• To improve the accuracy and efficiency of medical image analysis in diagnosing lung conditions.

Main Methods:

• Developed MixNet-LD, an automated system using a pre-trained MixNet model.
• Implemented a pre-processing strategy with Grad-Cam for noise reduction and feature highlighting.
• Utilized data augmentation for dataset balancing and employed dense blocks for improved classification.
• Classified images into normal, COVID-19, pneumonia, tuberculosis, and lung cancer categories using an SVM classifier.

Main Results:

• MixNet-LD achieved state-of-the-art performance with manageable model complexity.
• The system demonstrated high accuracy, reaching 98.5% on a challenging lung disease dataset.
• Tested on diverse datasets including the novel Pak-Lungs dataset.

Conclusions:

• MixNet-LD effectively enhances classification accuracy in medical image analysis.
• The proposed approach offers improved performance and learning capabilities for lung disease detection.
• This research contributes to developing advanced medical image processing strategies for clinical applications.