Simultaneous Tissue Classification and Lateral Ventricle Segmentation via a 2D U-net Driven by a 3D Fully...

Area of Science:

• Medical Image Analysis
• Neuroimaging
• Artificial Intelligence in Medicine

Background:

• Accurate segmentation of brain tissues and structures is crucial for neurological research and clinical diagnosis.
• Existing methods for brain segmentation often require manual intervention or lack comprehensive accuracy across multiple tissue types.

Purpose of the Study:

• To develop and validate a novel, fully automatic pipeline for simultaneous tissue classification and lateral ventricle segmentation.
• To enhance the accuracy and efficiency of brain structure segmentation in medical imaging.

Main Methods:

• A hybrid deep learning architecture combining a 3D fully convolutional neural network (FCN) for initial probability map generation and a 2D U-net for final segmentation.
• Utilized T1-weighted MRI atlases pre-segmented into gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), lateral ventricles (LVs), skull, and background.
• Processed images in a 2D slice fashion to incorporate global image context.

Main Results:

• The proposed pipeline demonstrated superior segmentation performance for GM, WM, CSF, LVs, and skull compared to a 3D patch-based FCN.
• Outperformed classical methods like SPM and FSL in segmenting GM and WM.
• Achieved accurate simultaneous classification and segmentation of multiple brain regions.

Conclusions:

• The developed automatic pipeline offers a robust and accurate solution for brain tissue and lateral ventricle segmentation.
• The proposed segmentation architecture is versatile and can be extended to various other medical image segmentation tasks.
• This approach holds significant potential for advancing neuroimaging analysis and applications.