Classifying Pneumonia among Chest X-Rays Using Transfer Learning.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Chest radiography is the primary method for pneumonia diagnosis.
• Analyzing chest X-rays requires expertise and can be time-consuming, especially in resource-limited areas.
• Computer-aided diagnosis systems are needed to overcome these limitations.

Purpose of the Study:

• To investigate the effectiveness of deep transfer learning for classifying pneumonia in chest X-ray images.
• To compare the performance of deep transfer learning against training deep learning models from scratch for pneumonia detection.

Main Methods:

• Three deep learning models (ResNet-50, Inception V3, DenseNet121) were trained using both deep transfer learning and from scratch.
• Models were fine-tuned slightly when using transfer learning.
• Performance was evaluated using the area under the curve (AUC).

Main Results:

• Deep transfer learning models achieved a 4.1% to 52.5% larger AUC compared to models trained from scratch.
• Slight fine-tuning of pre-trained architectures demonstrated a performance advantage.
• Transfer learning proved effective for pneumonia classification in chest X-rays.

Conclusions:

• Deep transfer learning is an effective strategy for improving pneumonia classification accuracy on chest X-ray images.
• This method offers a reduced development cost compared to training deep networks from scratch.
• The findings support the use of deep transfer learning in medical imaging for conditions like pneumonia.