Robust MR image segmentation using the trimmed likelihood estimator in asymmetric Student's-t mixture model

Area of Science:

• Medical Imaging
• Computer Vision
• Statistical Modeling

Background:

• Finite mixture models (FMM) are common for unsupervised segmentation of MR images.
• Real-world MR image data often contains outliers that negatively impact FMM parameter estimation.
• Robust statistical methods are needed to address outlier interference in image segmentation.

Purpose of the Study:

• To propose a robust estimation method for asymmetric Student's-t mixture models (ASMM) for MR image segmentation.
• To develop a technique that effectively discards outliers before parameter estimation.
• To enhance the accuracy and reliability of MR image segmentation in the presence of data imperfections.

Main Methods:

• Utilized a trimmed likelihood estimator for robust parameter estimation within the ASMM framework.
• Implemented an expectation-maximization (EM) algorithm to optimize the log-likelihood function.
• Applied the proposed method to segment synthetic datasets and real MR images.

Main Results:

• The proposed ASMM with trimmed likelihood estimation demonstrated robustness against outliers.
• The method showed improved performance in segmenting both synthetic and real MR images compared to standard approaches.
• The algorithm effectively discards outlier data points, leading to more accurate parameter estimation.

Conclusions:

• The robust ASMM with trimmed likelihood estimation offers a superior approach for MR image segmentation.
• This method provides flexibility in modeling data distributions and resilience to outliers.
• The findings suggest significant potential for this technique in clinical MR image analysis.