Specialized gray matter segmentation via a generative adversarial network: application on brain white matter...

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• White matter hyperintensities (WMH) on FLAIR MRI are key indicators in neurodegenerative diseases like Multiple Sclerosis (MS).
• Current automated WMH classification systems are lacking, necessitating advanced detection and classification methods for improved diagnosis and monitoring.
• Juxtacortical WMH (JCWMH) presents unique classification challenges due to difficulties in accurately segmenting gray matter (GM) masks.

Purpose of the Study:

• To develop an automated method for specialized gray matter (GM) segmentation for improved WMH classification, especially JCWMH.
• To create a pipeline integrating various image masks to refine GM maps and train a conditional generative adversarial network (cGAN).
• To enable WMH classification based on distances to ventricular and GM masks, reducing reliance on multi-sequence MRI inputs.

Main Methods:

• A novel pipeline was developed to integrate white matter, cerebrospinal fluid, ventricles, and WMH masks for refining GM maps.
• A conditional generative adversarial network (cGAN) was trained using the pipeline to enable classification using only FLAIR images.
• WMH classification was performed by analyzing distances between WMH and ventricular/GM masks, utilizing a manually segmented local dataset.

Main Results:

• The proposed method achieved a Dice similarity coefficient of 0.76, precision of 0.69, and sensitivity of 0.84 for JCWMH classification.
• It significantly outperformed the conventional approach using T1-derived GM masks, which yielded lower scores (0.66, 0.55, 0.81).
• The automated method processed images in under one second, compared to over two minutes for traditional methods.

Conclusions:

• The developed method offers an automated, fast, and initialization-free approach for WMH analysis exclusively using FLAIR images.
• This innovation simplifies complexity and enhances efficiency, facilitating more accurate clinical analysis of WMH.
• The findings pave the way for improved diagnostic and monitoring tools for brain diseases characterized by WMH.