AM-UNet

AM-UNet performs automated segmentation of the human brain claustrum from T1/T2 magnetic resonance imaging (MRI) data using a mini 3D U-Net-based deep learning architecture.

Key Features:

• Mini 3D U-Net Architecture: Implements a lightweight 3D U-Net-based network optimized for segmentation of the thin, sheet-like claustrum structure in MRI data.
• End-to-End Segmentation Pipeline: Integrates preprocessing, segmentation, and post-processing within a unified automated framework.
• Multi-Modal MRI Support: Processes combined T1/T2 MRI datasets for claustrum segmentation.
• High Segmentation Performance: Achieves Dice score of 82%, Intersection over Union (IoU) of 70%, and Intraclass Correlation Coefficient (ICC) of 90% on a T1/T2 claustrum MRI dataset.

Scientific Applications:

• Neuroimaging Structure Segmentation: Enables automated segmentation of the brain claustrum in magnetic resonance imaging studies.
• Brain Structure Analysis: Supports research investigating the anatomical and functional roles of the claustrum in the human brain.
• Neuroimaging Dataset Processing: Facilitates large-scale analysis of MRI datasets requiring precise segmentation of thin brain structures.

Methodology:

AM-UNet applies a mini 3D U-Net deep learning architecture to T1/T2 MRI data and integrates preprocessing and post-processing steps within an end-to-end segmentation framework to delineate the brain claustrum.