Multi-atlas Segmentation Enables Robust Multi-contrast MRI Spleen Segmentation for Splenomegaly

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Non-invasive spleen volume estimation is vital for diagnosing splenomegaly.
• Magnetic Resonance Imaging (MRI) is used for in vivo diagnosis, but accurate spleen volume estimation is challenging due to anatomical variations and diverse imaging modalities.
• Multi-atlas segmentation offers a robust approach for segmenting heterogeneous medical image data.

Purpose of the Study:

• To propose and evaluate multi-atlas segmentation frameworks for spleen segmentation in MRI scans for splenomegaly detection.
• To introduce novel automated and semi-automated atlas selection methods tailored for spleen MRI.

Main Methods:

• Implementation of multi-atlas segmentation for MRI spleen segmentation.
• Development of an automated atlas selection method using the Selective and Iterative Method for Performance Level Estimation (SIMPLE) approach.
• Introduction of a semi-automated method (L-SIMPLE) incorporating craniocaudal length priors to guide atlas selection and control outliers.

Main Results:

• Both automated and semi-automated multi-atlas methods achieved a median Dice Similarity Coefficient (DSC) greater than 0.9.
• The L-SIMPLE method effectively reduced outliers with minimal manual effort (approximately 1 minute per scan).
• L-SIMPLE achieved a high Pearson correlation of 0.9713 when compared to manual spleen segmentation.

Conclusions:

• Multi-atlas segmentation is a highly effective technique for accurate spleen segmentation in multi-contrast MRI scans for splenomegaly.
• The proposed L-SIMPLE method provides an efficient and accurate semi-automated approach for spleen segmentation, improving upon automated methods by controlling outliers.
• This approach holds significant potential for improving the clinical diagnosis and monitoring of splenomegaly using MRI.