iSTAPLE: Improved Label Fusion for Segmentation by Combining STAPLE with Image Intensity

Area of Science:

• Medical image analysis
• Computational anatomy
• Computer-aided diagnosis

Background:

• Multi-atlas segmentation methods are widely used for automated image segmentation.
• Current methods like STAPLE fuse propagated label maps but neglect target image intensity, limiting accuracy.
• Existing techniques often require similar image contrasts and intensity ranges between target images and atlases.

Purpose of the Study:

• To develop an improved multi-atlas segmentation method, iSTAPLE, that incorporates target image intensity.
• To enhance segmentation accuracy by combining intensity-based information with atlas consensus.
• To overcome limitations of existing methods regarding image contrast and intensity range variability.

Main Methods:

• Developed iSTAPLE, a novel method integrating target image intensity into a maximum likelihood estimate (MLE) framework.
• Utilized a modified Expectation-Maximization (EM) algorithm to simultaneously estimate intensity profiles, true segmentation, and atlas performance.
• Designed iSTAPLE to be independent of target image contrast and intensity range compared to atlases.

Main Results:

• iSTAPLE demonstrated superior performance compared to STAPLE in whole brain segmentation experiments.
• The method effectively leverages both intensity information and atlas-based consensus for improved segmentation.
• Simultaneous estimation of intensity profiles, segmentation, and atlas performance was achieved.

Conclusions:

• iSTAPLE offers a more robust and accurate approach to multi-atlas image segmentation.
• The method's ability to handle varying image contrasts and intensities broadens its applicability.
• iSTAPLE represents a significant advancement in automated medical image segmentation techniques.