RIL-Contour: a Medical Imaging Dataset Annotation Tool for and with Deep Learning

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Deep learning (DL) models require extensive annotated datasets for optimal performance.
• Manual curation of these datasets is a significant bottleneck in DL development.
• Existing annotation tools often lack integration with DL model training workflows.

Purpose of the Study:

• To develop RIL-Contour, a software tool to accelerate medical image annotation for and with deep learning.
• To enable clinical users to leverage DL models for rapid medical image annotation.
• To facilitate collaboration between image analysts, radiologists, and data scientists.

Main Methods:

• RIL-Contour supports manual, semi-automated, and fully automated annotation methods.
• The software implements Annotation by Iterative Deep Learning (AID) for accelerated annotation.
• It standardizes annotations to reduce errors and promotes collaborative workflows.
• Mechanisms for automated feedback loops between data scientists and image analysts are included.

Main Results:

• RIL-Contour significantly accelerates the process of medical image annotation.
• The AID methodology enables iterative annotation, training, and utilization of DL models.
• The software facilitates seamless collaboration, improving efficiency in dataset curation and model development.
• Standardized annotations reduce errors and enhance dataset quality.

Conclusions:

• RIL-Contour effectively addresses the challenge of medical image dataset curation for deep learning.
• The AID methodology and collaborative features expedite DL model development in clinical settings.
• RIL-Contour empowers clinically oriented users to utilize DL for rapid annotation and analysis.