Proactive Construction of an Annotated Imaging Database for Artificial Intelligence Training.

Area of Science:

• Medical Imaging Informatics
• Artificial Intelligence in Healthcare
• Oncology Diagnostics

Background:

• Artificial intelligence (AI) shows significant potential for enhancing medical imaging diagnostics.
• A major limitation in developing clinical-grade AI is the scarcity of large, high-quality annotated training datasets.
• High-quality ground truth annotation is crucial for effective AI model development in medical imaging.

Purpose of the Study:

• To establish and detail the construction of a comprehensive database of annotated oncology imaging data.
• To support diverse AI training tasks including detection, quantification, segmentation, and classification.
• To focus on annotation quality and generality for broad AI applicability.

Main Methods:

• Compilation of a labeled image dataset encompassing oncology imaging from pathology and radiology.
• Inclusion of data from multiple cancer types: breast, ovary, skin, colon, skeleton, and liver.
• Exploration of best practices for scalable, high-quality medical image collection.

Main Results:

• Successful creation of a detailed, annotated oncology imaging database.
• The database contains labeled image data across six distinct anatomical regions and cancer types.
• Identification of generic lessons learned for constructing medical imaging databases for AI training.

Conclusions:

• The developed database serves as a valuable resource for AI model development in oncology.
• Eight guiding principles for organizing efforts in medical imaging database construction were summarized.
• This work provides a framework for scalable, high-quality medical image data collection for AI.