COVID-19-CT-CXR

COVID-19-CT-CXR provides a public database of chest X‑ray (CXR) and computed tomography (CT) images extracted from the PubMed Central Open Access (PMC-OA) Subset to support medical image analysis and diagnostic model development.

Key Features:

• Automated extraction: Extracts figures, associated captions, and detailed figure descriptions from PMC-OA articles and separates compound figures into individual subfigures.
• Deep-learning classification: Applies a deep-learning model to distinguish CXR and CT images from other figure types in publications.
• Text-mined contextual data: Associates each image with text-mined captions and figure descriptions that provide fine-grained clinical details.
• Image collection statistics: Includes 1,327 CT images and 263 CXR images (as of May 9, 2020).
• Complementary data modality: Provides paired image and textual data intended to complement existing medical imaging datasets for analysis.

Scientific Applications:

• Deep-learning training: Serves as additional training data to improve classification performance for COVID-19 versus non-COVID-19 CT scans.
• Differential diagnosis: Includes CT images of influenza to aid neural networks in distinguishing COVID-19, influenza, and other diseases on CT.
• Anomaly detection: Supports unsupervised one-class classifier approaches trained on non-COVID-19 CXR data for detecting COVID-19 anomalies.
• Clinical symptom comparison: Enables comparative analyses of clinical findings between COVID-19 and influenza using text-mined captions and figure descriptions.

Methodology:

Systematic extraction from the PubMed Central Open Access (PMC-OA) Subset with extraction of figures, captions, and figure descriptions, decomposition of compound figures into subfigures, application of a deep-learning model to identify/classify CXR and CT images, and use of an unsupervised one-class classifier on non-COVID-19 CXR data for anomaly detection.