IntellEditS: intelligent learning-based editor of segmentations.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Image Processing

Background:

• Automatic segmentation methods often yield local inaccuracies in 3D datasets.
• Manual correction of these segmentations is time-consuming and labor-intensive.
• Improving segmentation accuracy and reducing correction effort are critical for clinical applications.

Purpose of the Study:

• To introduce Intelligent Learning-Based Editor of Segmentations (IntellEditS), a novel tool to minimize user effort in correcting segmentation errors.
• To enhance the accuracy of 3D image segmentations through an interactive, learning-based approach.
• To provide a versatile tool capable of handling various segmentation representations and user interactions.

Main Methods:

• IntellEditS integrates interactive learning with an energy-minimization framework.
• A discriminative classifier is trained on user input for soft voxel labeling within edited 3D regions.
• The system incorporates these labels into a novel energy functional alongside original segmentation and image data.
• The tool supports correction of both segmentation masks and meshes, accepting intuitive boundary-based user input.

Main Results:

• IntellEditS effectively minimizes user effort required for segmentation correction.
• The tool demonstrates improved segmentation accuracy compared to existing methods.
• Successful application and validation were shown on diverse MRI and CT datasets with varied anatomical structures and resolutions.

Conclusions:

• IntellEditS offers a powerful and efficient solution for correcting segmentation inaccuracies in 3D medical imaging.
• The tool's ability to handle masks and meshes, coupled with intuitive interactions, enhances its clinical utility.
• This approach represents a significant advancement in automated image segmentation refinement.