CMed: Crowd Analytics for Medical Imaging Data.

Area of Science:

• Medical image analysis
• Data visualization
• Crowdsourcing

Background:

• Crowdsourcing is increasingly used for medical image annotation.
• Analyzing the quality and patterns of crowdsourced annotations is crucial for effective application design.
• Existing tools may lack specialized features for exploring diverse annotation metrics.

Purpose of the Study:

• To introduce CMed, a visual analytics framework for medical image data annotations from crowdsourcing.
• To enable visualization, classification, and filtering of crowdsourced clinical data using various metrics.
• To assist crowdsourcing application analysts in understanding patterns and optimizing future applications.

Main Methods:

• Developed CMed, a framework with interactive linked visualization components.
• Utilized metrics such as detection rate, logged events, and annotation clustering for analysis.
• Applied the framework to polyp detection in virtual colonoscopy and lung nodule detection in CT images.

Main Results:

• CMed allows detailed inspection of individual worker results through multiple linked views.
• The framework facilitates observation of patterns and insights within crowdsourced medical data.
• Demonstrated efficacy through two distinct medical crowdsourcing studies.

Conclusions:

• CMed provides a robust platform for analyzing crowdsourced medical image annotations.
• The framework aids in designing more effective crowdsourcing applications for medical data.
• Lessons learned offer insights for integrating such frameworks into clinical workflows.