Segmentation Labeling Tool

A command-line tool for comparing and labeling the quality of two different segmentation methods on medical images. The tool displays segmentations side-by-side using Freeview and collects structured feedback about their relative quality.

Features

• Interactive comparison of two segmentation methods
• Randomized presentation to prevent bias
• Automatic focus on areas with significant differences
• Structured feedback collection including:
  • Better performing method
  • Confidence level
  • Magnitude of differences
  • Segmentation failure detection
  • Custom comments

Usage

python labeling_tool.py --method1 <method1> --method2 <method2> [options]

Required Arguments

• -m1, --method1: Name of the first segmentation method
• -m2, --method2: Name of the second segmentation method
• -o, --output_file: Path to save the CSV results
• -i, --input_data: CSV file containing input data with columns: "subjectID", "image", "method1", "method2"
• --diff_maps_dir: Directory containing difference maps

Optional Arguments

• --user: Name of the labeler (defaults to system username)
• --fs: Path to FreeSurfer home directory

Output

Results are saved in CSV format with the following information:

• Subject ID
• Method names
• Better performing method
• Confidence level (random/uncertain/certain)
• Difference strength (none/marginal/moderate/substantial)
• Segmentation failures
• User comments
• Labeler name
• Time taken
• Number of differences

Requirements

• FreeSurfer
• Python 3.x
• nibabel
• pandas
• tqdm
• xdotool (optional, for window management)