Robust Multicontrast MRI Spleen Segmentation for Splenomegaly Using Multi-Atlas Segmentation

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Spleen volume measurement from 3D MRI for splenomegaly diagnosis is challenging due to anatomical variations and diverse acquisition schemes.
• Multi-atlas segmentation (MAS) is a validated approach for handling anatomical heterogeneity in medical imaging.

Purpose of the Study:

• To propose and evaluate MAS for spleen segmentation in clinical MRI for splenomegaly.
• To improve the accuracy and reliability of spleen volume measurements in patients with splenomegaly.

Main Methods:

• Implemented an automated segmentation method using selective and iterative method for performance level estimation (SIMPLE) atlas selection.
• Developed a semiautomated L-SIMPLE method integrating spatial priors for outlier control.
• Employed graph cuts refinement for final segmentation from MAS probability maps.

Main Results:

• Evaluated methods on 55 clinical MRI volumes (28 T1-weighted, 27 T2-weighted).
• Both automated and semiautomated methods achieved high segmentation accuracy (median Dice coefficient).
• L-SIMPLE significantly alleviated outliers and achieved 0.97 Pearson correlation with manual volume measurements, requiring minimal manual effort.

Conclusions:

• MAS is effective for spleen segmentation in clinical MRI for splenomegaly.
• The L-SIMPLE method offers an improved approach for accurate spleen volume quantification.
• This technique aids in noninvasive diagnosis and monitoring of splenomegaly.