Uncertainty-aware guided volume segmentation.

Area of Science:

• Medical imaging
• Computer-assisted diagnosis
• Scientific visualization

Background:

• Direct volume rendering is crucial for volumetric data but lacks efficient and reliable feature detection.
• Current methods often involve a trade-off between speed and accuracy, neglecting segmentation uncertainty.
• Existing approaches typically overlook the uncertainty inherent in feature detection processes.

Purpose of the Study:

• To propose a guided probabilistic volume segmentation approach that minimizes uncertainty.
• To enhance the reliability and efficiency of feature detection in volumetric data.
• To address the neglected issue of uncertainty in feature detection.

Main Methods:

• An iterative process assessing uncertainty of random walker-based segmentation.
• Directing user attention to ambiguous regions for classification correction.
• Incorporating user-provided classification information to improve efficiency.
• GPU implementation using OpenCL API for interactive workflow.

Main Results:

• Demonstrated reliability and efficiency across various medical datasets (brain MRI, abdominal CT).
• Reduced risk of critical segmentation errors by focusing on high-ambiguity regions.
• Provided dependable information on segmentation reliability to the user.

Conclusions:

• The guided probabilistic segmentation approach effectively minimizes uncertainty.
• The method offers a reliable and efficient solution for feature detection in medical imaging.
• Interactive user guidance significantly improves segmentation accuracy and trustworthiness.