COVID-19 classification of X-ray images using deep neural networks

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Disease Detection
• Radiology and Diagnostic Imaging

Background:

• Chest X-ray (CXR) imaging is crucial for diagnosing and monitoring COVID-19.
• The COVID-19 pandemic necessitated rapid advancements in diagnostic tools.
• Efficient patient management requires accurate and timely interpretation of medical images.

Purpose of the Study:

• To develop a deep learning model for detecting COVID-19 from CXRs.
• To create a tool for retrieving similar patient CXR images based on model analysis.
• To assist medical professionals in managing COVID-19 cases and reduce workload.

Main Methods:

• A retrospective study utilized 1384 COVID-19 positive CXRs and 1024 non-COVID CXRs.
• An ensemble of deep neural networks (ResNet34, ResNet50, ResNet152, VGG16) was employed.
• Data augmentation, lung segmentation, and a nearest-neighbors algorithm were incorporated.

Main Results:

• The model achieved 90.3% accuracy, 90% specificity, and 90.5% sensitivity.
• The Area Under the Receiver Operating Characteristic Curve (AUC) was 0.96.
• A tool was developed to find similar CXR images using DNN-based embeddings.

Conclusions:

• Deep learning models trained on CXRs can effectively detect COVID-19.
• The developed AI tools can support medical efforts and alleviate staff burden.
• The system demonstrates high performance in identifying COVID-19 positive cases from chest X-rays.