Efficient multi-atlas abdominal segmentation on clinically acquired CT with SIMPLE context learning.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Computational Anatomy

Background:

• Abdominal segmentation in computed tomography (CT) is complex due to anatomical variations and intricate organ relationships.
• Multi-atlas segmentation (MAS) offers a robust solution by utilizing registered label atlases and statistical fusion.
• Existing MAS methods require optimization, particularly in atlas selection, to address registration errors.

Purpose of the Study:

• To develop and evaluate an improved MAS technique for segmenting 12 abdominal structures in clinically acquired CT scans.
• To enhance atlas selection and label fusion strategies within the MAS framework.
• To investigate the impact of context learning and advanced fusion methods on segmentation accuracy.

Main Methods:

• A re-derived Selective and Iterative Method for Performance Level Estimation (SIMPLE) algorithm was employed, incorporating Bayesian priors for context learning.
• Joint Label Fusion (JLF) was integrated to mitigate correlated errors from selected atlases.
• Graph cut techniques were utilized for segmentation regularization.

Main Results:

• The proposed method demonstrated superior performance compared to existing MAS techniques (majority vote, SIMPLE, JLF, Wolz).
• Significant Dice Similarity Coefficient (DSC) improvements were observed: a median of 7.0% over JLF and 16.2% over Wolz.
• The approach achieved consistent improvements across 100 subjects, highlighting its effectiveness.

Conclusions:

• The novel MAS technique, integrating context learning and advanced fusion, significantly enhances abdominal organ segmentation accuracy in CT.
• This method offers a more efficient and robust solution for large-scale CT data analysis in clinical applications.
• The findings support the use of this technique for applications like biomarker screening, surgical navigation, and data mining.