Deep Generative Classifiers for Thoracic Disease Diagnosis with Chest X-ray Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Thoracic diseases pose significant health risks, necessitating accurate diagnostic tools.
• Chest X-ray interpretation is crucial but challenging for radiologists.
• Existing deep learning models for chest X-ray classification are often susceptible to noise and overfitting.

Purpose of the Study:

• To introduce deep generative classifiers for robust and accurate thoracic disease diagnosis from chest X-ray images.
• To address the limitations of deterministic deep learning architectures in handling noise and overfitting.

Main Methods:

• Proposed deep generative classifiers featuring a distribution middle layer and a sampling layer.
• Implemented generative classifiers based on established deterministic neural network architectures.
• Evaluated models on the chest X-ray14 dataset.

Main Results:

• Deep generative classifiers demonstrated superior performance compared to their deterministic counterparts.
• The proposed models showed increased robustness to noise.
• Reduced overfitting was observed in the generative deep learning approach.

Conclusions:

• Deep generative classifiers offer a promising advancement for automated thoracic disease diagnosis.
• These models provide a more robust and accurate alternative to traditional deep learning methods for medical image analysis.
• The findings highlight the potential of generative approaches in overcoming common challenges in AI-driven medical diagnostics.