ThoraciNet: thoracic abnormality detection and disease classification using fusion DCNNs

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-rays are crucial for diagnosing thoracic abnormalities but analysis is complex and requires expert radiologists.
• Asymptomatic diseases pose diagnostic challenges, highlighting the need for advanced analytical tools.
• Large-scale, annotated chest X-ray datasets facilitate the development of AI-powered diagnostic systems.

Purpose of the Study:

• To develop and evaluate deep learning models for automating thoracic abnormality detection, classification, and segmentation from chest X-rays.
• To propose a two-stage pipeline integrating deep convolutional neural networks for enhanced diagnostic performance.
• To improve the accuracy and interpretability of computer-aided diagnosis systems in radiology.

Main Methods:

• A two-stage deep learning pipeline was implemented for abnormality detection and disease classification.
• Two fusion-based models utilizing asymmetric deep convolutional neural networks were developed for classification.
• Model performance was evaluated on the NIH chest X-ray database using accuracy and AUC scores.

Main Results:

• The proposed deep learning architecture achieved an AUC score of 0.99 for chest X-ray (CXR) triaging.
• An AUC score of 0.79 was obtained for CXR disease classification.
• GradCAM visualization confirmed the interpretability of model predictions for clinical validation.

Conclusions:

• Deep learning models offer a promising approach to automate and enhance the accuracy of chest X-ray analysis.
• The proposed fusion-based models demonstrate superior performance compared to existing methods.
• AI-driven tools can assist radiologists, improving diagnostic efficiency and patient outcomes.