Classification and visual explanation for COVID-19 pneumonia from CT images using triple learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Standard deep learning classification methods struggle with medical images due to variations in disease progression and lesion size.
• Transparency and explainability are crucial for healthcare providers to trust AI models in medical diagnostics.
• Accurate classification and visualization of COVID-19 pneumonia from CT images are essential for timely diagnosis and treatment.

Purpose of the Study:

• To develop a novel framework for classifying and visualizing COVID-19 pneumonia using CT images.
• To address the limitations of conventional methods in handling variations in medical image data.
• To enhance model transparency and explainability for clinical trust.

Main Methods:

• Utilized contrastive learning to improve feature representation and classification accuracy by minimizing distance between similar images.
• Integrated an attention mechanism to emphasize critical regions within CT images, aiding in classification and visualization.
• Employed a three-fold cross-validation strategy for robust experimental evaluation.

Main Results:

• Achieved significant improvements in classification accuracy compared to conventional methods.
• Demonstrated the ability to provide detailed visual explanations of the classification process.
• Validated the framework's effectiveness through experiments on two distinct classification tasks.

Conclusions:

• The proposed framework effectively classifies and visualizes COVID-19 pneumonia from CT images.
• Contrastive learning and attention mechanisms enhance both accuracy and interpretability in medical image analysis.
• This novel approach offers a reliable and transparent tool for radiologists and clinicians.