RIL-Contour

RIL-Contour enables annotation and iterative refinement of medical imaging datasets to produce standardized, high-quality labels for training supervised deep-learning models.

Key Features:

• Support for Multiple Annotation Methods: Supports fully automated, semi-automated, and manual annotations, including voxel and text annotations.
• Annotation by Iterative Deep Learning (AID): Implements AID as iterative cycles of annotating images, training deep-learning models, and using trained models to refine annotations.
• Standardization of Annotations: Enforces standardized annotations across datasets to reduce annotation errors and improve model robustness.
• Collaborative Workflow Support: Enables role-based collaboration where image analysts annotate, radiologists approve clinical labels, and data scientists train models using validated annotations.
• Automated Feedback Loop: Provides an automated feedback loop that returns newly trained deep-learning models to annotators for continuous dataset refinement.

Scientific Applications:

• Medical imaging dataset curation: Produces large, standardized labeled datasets for training supervised deep-learning models in medical imaging research.
• Clinical decision-support development: Supports development and validation of AI systems that assist clinical decision-making through standardized annotations and iterative model improvement.

Methodology:

Uses Annotation by Iterative Deep Learning (AID): iterative cycles of annotating images, training deep-learning models on those annotations, applying models to refine annotations, and incorporating expert (radiologist) approvals.