AI-Driven Thoracic X-ray Diagnostics: Transformative Transfer Learning for Clinical Validation in Pulmonary...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Pulmonary radiography is crucial for diagnosing thoracic conditions.
• Current diagnostic methods face challenges in accuracy and efficiency.
• Advanced artificial intelligence (AI) offers potential solutions.

Purpose of the Study:

• To evaluate advanced AI methodologies for enhancing diagnostic accuracy in pulmonary radiography.
• To assess the performance of AI models compared to expert radiologists.
• To explore the integration of AI and natural language processing (NLP) in clinical workflows.

Main Methods:

• Utilized DenseNet121 and ResNet50 on 108,948 chest X-ray images.
• Applied Latent Dirichlet Allocation (LDA) and Named Entity Recognition (NER) for clinical text analysis.
• Employed DistilBERT for sentiment analysis and XGBoost with SHAP for interpretability.
• Incorporated Local Interpretable Model-agnostic Explanations (LIME) and occlusion sensitivity analysis.

Main Results:

• DenseNet121 achieved a 94% AUC in identifying pneumothorax and edema.
• AI models surpassed expert radiologists in specific diagnostic tasks.
• NER system achieved 92% precision and 88% recall.
• AI techniques reduced processing times by 60% and annotation errors by 75%.

Conclusions:

• AI demonstrates significant potential to advance thoracic diagnostics and accelerate medical evaluations.
• AI integration enhances efficiency and accuracy in pulmonary radiography.
• Further research is needed for complex condition diagnosis, but AI shows transformative potential in medical imaging.