Content-based medical image retrieval system for lung diseases using deep CNNs

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Content-based image retrieval (CBIR) systems analyze latent image features for relevance, bypassing textual descriptions.
• Traditional CBIR relies on feature vectors for similarity matching.
• Accurate retrieval of medical images is crucial for diagnostics.

Purpose of the Study:

• To propose a content-based medical image retrieval (CBMIR) system for early lung disease detection and classification.
• To leverage deep neural models and transfer learning for analyzing lung X-ray images.
• To enhance the accuracy of identifying disease-specific features in medical imaging.

Main Methods:

• Developed a CBMIR system utilizing deep neural networks.
• Employed transfer learning on COVID-19 Chest X-ray datasets for model training.
• Evaluated the system's performance on a standard medical image dataset.

Main Results:

• Achieved a 49.71% improvement in precision across various distance metrics.
• Observed a 26.55% improvement in the area under the precision-recall curve (AUPRC).
• Demonstrated enhanced identification and classification of disease-specific features.

Conclusions:

• The proposed CBMIR system shows significant potential for early lung disease detection.
• Deep learning and transfer learning are effective for analyzing medical images in CBMIR.
• The system offers a substantial improvement in retrieval and classification performance for lung X-rays.