Clinically validated dataset of 435 human colons segmented from CT colonography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Anatomy

Background:

• High-quality segmentation datasets are crucial for AI in medical imaging.
• Segmenting complex organs like the colon presents significant challenges.
• Existing datasets lack comprehensive colon segmentation for AI development.

Purpose of the Study:

• To create the first open-access, clinically validated dataset of segmented human colons from CTC.
• To support AI-driven advancements in medical image analysis for colon-related conditions.
• To provide a resource for population-scale radiologic studies and anatomical modeling.

Main Methods:

• Utilized 435 human colon scans from The Cancer Imaging Archive (TCIA).
• Performed segmentation of the whole colon, including collapsed segments and fluid.
• Generated masks for gas-filled colon regions.
• Ensured clinical validation of segmentation accuracy by an expert radiologist.

Main Results:

• Developed a dataset of 435 segmented human colons from CTC.
• Included whole colon masks and gas-filled colon masks for each scan.
• Achieved clinically validated segmentation accuracy.
• Established the first open-access resource of its kind.

Conclusions:

• The new dataset is a valuable resource for AI in medical imaging, particularly for colon analysis.
• Facilitates population-scale radiologic studies and AI tool development.
• Enables the creation of accurate digital models and simulators for research and training.