medical_chest_xray_classifier

Overview

This model is a ResNet-50 based image classifier fine-tuned on clinical chest radiography. It is designed to categorize X-ray images into three distinct classes: Normal, Bacterial Pneumonia, and Viral Pneumonia. The model assists in rapid triage and secondary diagnostic validation in clinical settings.

Model Architecture

The model utilizes a ResNet-50 architecture, leveraging deep residual learning to extract complex thoracic features.

• Backbone: 50 layers of convolutional blocks with shortcut connections to mitigate vanishing gradients.
• Input: Normalized 224x224 grayscale-converted images.
• Optimization: Trained using cross-entropy loss with a focus on maximizing the macro-averaged $F_1$ score to handle class imbalances.

Intended Use

• Radiology Triage: Prioritizing scans that show high probabilities of infection for immediate human review.
• Medical Education: Aiding students in identifying visual markers of different pneumonia types.
• Research: Serving as a baseline for more complex multi-modal diagnostic systems.

Limitations

• Diagnostic Tool Only: This model is a supportive tool and must not be used for final medical diagnosis without professional oversight.
• Hardware Variation: Performance may vary based on the manufacturer of the X-ray equipment and exposure settings.
• Demographics: Validation was primarily conducted on adult datasets; pediatric accuracy may be significantly lower.