LDD-VTS: an AI-based framework for lung disease diagnosis using vision transformers and SHAP

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pulmonology

Background:

• Lung diseases, especially lung cancer, present significant global health challenges with high mortality.
• Current diagnostic methods for lung diseases often lack the desired accuracy and timeliness.
• Early and precise diagnosis is critical for effective patient management and improved outcomes.

Purpose of the Study:

• To develop and evaluate a novel AI framework (LDD-VTS) for enhanced diagnosis and classification of lung diseases.
• To utilize Vision Transformers (ViTs) and SHAP for improved accuracy and interpretability in lung disease detection.
• To process medical imaging data (X-rays, CT scans) for accurate identification of lung abnormalities.

Main Methods:

• Implementation of a novel framework, LDD-VTS, integrating Vision Transformers (ViTs) and SHAP (SHapley Additive exPlanations).
• Processing of diverse medical imaging data, including chest X-rays and CT scans, for lung disease analysis.
• Utilizing the P16-224-In21K model configuration for classification tasks.

Main Results:

• The P16-224-In21K model configuration achieved 98.43% accuracy on the IQ-OTH/NCCD lung cancer dataset.
• The framework demonstrated high precision and recall in identifying lung cancer, viral pneumonia, and lung opacity.
• SHAP integration provided transparent insights into model predictions by highlighting influential image regions.

Conclusions:

• The LDD-VTS framework shows significant potential for transforming lung disease diagnostics through AI.
• The combination of ViTs and SHAP enhances diagnostic accuracy and clinical interpretability.
• This approach facilitates informed decision-making, promoting earlier detection and personalized treatment strategies for lung conditions.