The 2021 SIIM-FISABIO-RSNA Machine Learning COVID-19 Challenge: Annotation and Standard Exam Classification of...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate detection of COVID-19 on chest radiographs is crucial for patient management.
• Existing datasets require further curation and annotation for robust AI model training.
• Standardized annotation guidelines are needed for consistent interpretation of radiographic findings.

Purpose of the Study:

• To describe the creation and characteristics of a dataset for an artificial intelligence (AI) challenge.
• To facilitate the development and validation of AI algorithms for COVID-19 detection and localization on chest radiographs.
• To provide a valuable resource for researchers investigating AI applications in medical imaging for infectious diseases.

Main Methods:

• Dataset curation and annotation methodology for chest radiographs.
• International radiologist collaboration for image annotation.
• Development of mutually exclusive categories for COVID-19 appearance on radiographs: typical, indeterminate, atypical, and negative for pneumonia.
• Inclusion of bounding boxes for airspace opacities.

Main Results:

• A comprehensive dataset of chest radiographs annotated for COVID-19 detection.
• Standardized annotations based on established guidelines.
• Availability of bounding box data for precise lesion localization.
• Dataset characteristics suitable for AI challenge development.

Conclusions:

• The described dataset and annotation methodology provide a robust foundation for AI-driven COVID-19 detection.
• This resource supports academic and noncommercial research in medical artificial intelligence.
• The dataset facilitates advancements in automated analysis of chest radiographs for pneumonia and COVID-19.