Fully Automatic White Matter Hyperintensity Segmentation using U-net and Skip Connection.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in neuroscience
• Neurodegenerative disease research

Background:

• White matter hyperintensity (WMH) is a common finding in aging and neurodegenerative conditions.
• Accurate segmentation of WMH is crucial for understanding disease progression and evaluating treatments.
• Current segmentation methods may lack efficiency or accuracy across different imaging datasets.

Purpose of the Study:

• To develop and validate a fully automatic system for segmenting WMH.
• To introduce and evaluate a novel Skip Connection U-net (SC U-net) architecture.
• To compare the SC U-net's performance against a classical U-net for WMH segmentation.

Main Methods:

• Integration of classical image processing with deep neural networks.
• Development of a novel SC U-net architecture for enhanced segmentation.
• Experiments conducted on a dataset of 60 MRI scans from three different scanners, including cross-scanner validation.

Main Results:

• The proposed SC U-net demonstrated faster convergence compared to the classical U-net.
• The SC U-net achieved higher segmentation accuracy for WMH.
• The system's software and models were made publicly available via Dockerhub for accessibility.

Conclusions:

• The developed automatic system, particularly the SC U-net, provides an accurate and efficient tool for WMH segmentation.
• This advancement can aid in the clinical assessment and research of aging and neurodegenerative diseases.
• Public accessibility of the system promotes further research and application in the field.