Open-Access Fully Automated Intravenous Contrast Detection and Body Part Classification for Computed Tomography...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Intravenous contrast presence on computed tomography (CT) scans is often poorly documented in large datasets.
• Automated tools are needed to improve the efficiency and reliability of CT scan documentation.

Purpose of the Study:

• To develop and validate FALCON, an open-access deep learning model for automated intravenous contrast detection and body part classification in CT scans.
• To assess FALCON's performance and compare its annotation speed to human experts.

Main Methods:

• Trained and validated four convolutional neural network (CNN) models using ResNet9 architecture on six independent datasets (3138 CT scans).
• Verified ground truth for contrast presence by a radiologist.
• Integrated models into a graphical user interface for user-friendly operation.

Main Results:

• External test set (1348 scans) achieved high F1 scores for contrast detection: 99.4% (head and neck), 98.3% (chest), and 98.1% (abdomen/pelvis).
• Body part classification achieved 100% F1 score on unseen data.
• FALCON significantly outperformed human experts in annotation time (e.g., 1.3s vs. 21s for head and neck CT).

Conclusions:

• FALCON provides a fast and reliable automated solution for intravenous contrast detection and body part classification on CT scans.
• The open-access model has the potential to enhance large-scale research by improving data annotation efficiency and accuracy.