Pneumonia Detection from Chest X-Ray Images Using Deep Learning and Transfer Learning for Imbalanced Datasets.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Pneumonia is a major global health concern requiring prompt diagnosis.
• Deep learning offers automated pneumonia detection from chest X-rays.
• Imbalanced datasets present a significant challenge in developing robust AI models.

Purpose of the Study:

• To investigate deep learning for pneumonia detection, focusing on imbalanced datasets.
• To evaluate various deep learning architectures (VGG, ResNet, ViT) and mitigation strategies.
• To explore transfer learning, zero-shot, few-shot, and semi-supervised learning for improved performance.

Main Methods:

• Evaluation of VGG, ResNet, and ViT architectures on Chest X-Ray, BRAX, and CheXpert datasets.
• Application of transfer learning from ImageNet and data augmentation techniques.
• Implementation of semi-supervised learning (Mean Teacher) and balanced weight strategies.

Main Results:

• Transfer learning, data augmentation, and balanced weights significantly improved performance on imbalanced datasets.
• Deep learning models demonstrated high accuracy in pneumonia detection.
• Semi-supervised learning effectively utilized unlabeled data, enhancing diagnostic capabilities.

Conclusions:

• Deep learning, particularly with transfer learning and data balancing, shows great potential for accurate pneumonia detection.
• Strategy selection should be tailored to dataset characteristics.
• Semi-supervised learning offers a promising avenue for improving AI-driven medical diagnostics.