NextGen lung disease diagnosis with explainable artificial intelligence

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-rays (CXRs) are vital for diagnosing lung diseases like COVID-19, pneumonia, and tuberculosis.
• Manual interpretation of CXRs is difficult due to overlapping features, necessitating advanced diagnostic tools.
• Artificial Intelligence (AI) offers potential but lacks trust due to its 'black-box' nature.

Purpose of the Study:

• To develop a novel explainable AI (XAI) model, XAI-TRANS, for multiclass classification of lung diseases from CXR images.
• To address the challenge of overlapping radiological features in CXR interpretation.
• To enhance the trustworthiness of AI predictions in medical diagnostics.

Main Methods:

• Proposed a novel XAI-TRANS model utilizing inception-based transfer learning for CXR classification.
• Implemented an improved U-Net for lung segmentation to extract critical radiological features.
• Integrated explainable AI techniques (LIME and Grad-CAM) for model transparency.

Main Results:

• Achieved a maximum precision of 98% and accuracy of 97% in multiclass lung disease classification.
• Demonstrated a 4.75% improvement by leveraging XAI techniques for enhanced prediction explanations.
• Successfully transformed a 'black-box' AI into a transparent 'glass-box' model.

Conclusions:

• The XAI-TRANS model effectively classifies lung diseases from CXR images with high accuracy.
• Explainable AI significantly improves the reliability and interpretability of AI-driven medical diagnoses.
• This approach enhances diagnostic confidence and supports clinical decision-making in pulmonary disorder assessment.