Region detection by minimizing intraclass variance with geometric constraints, global optimality, and efficient...

Area of Science:

• Medical Image Analysis
• Computational Geometry
• Optimization

Background:

• Efficient segmentation of globally optimal surfaces is crucial for medical image analysis.
• Intraclass variance minimization is effective for object segmentation, particularly in images lacking sharp edges.

Purpose of the Study:

• To develop an efficient polynomial-time algorithm for segmenting regions between two coupled smooth surfaces by minimizing intraclass variance.
• To validate the algorithm's accuracy on synthetic and real-world datasets, including clinical data.

Main Methods:

• The algorithm utilizes shape probing techniques from computational geometry.
• It computes a sequence of minimum-cost closed sets within a derived parametric graph.
• An approximation algorithm is also presented for faster computation with comparable accuracy.

Main Results:

• The method demonstrated highly accurate segmentation results on computer-synthetic volumetric images and X-ray CT datasets.
• Applicability to clinical datasets was confirmed, showing the technique's broad utility.
• The approximation algorithm provided significantly faster performance with minimally reduced accuracy.

Conclusions:

• The developed technique offers an efficient and accurate solution for segmenting regions between surfaces in volumetric medical images.
• This approach has potential applications in various medical imaging optimization problems.
• Both the exact and approximation algorithms are valuable contributions to the field.