Improving brain atrophy quantification with deep learning from automated labels using tissue similarity priors.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Deep Learning

Background:

• Brain atrophy quantification via MRI is vital for neurodegenerative disease diagnosis and prognosis.
• Current methods face technical and biological challenges, limiting individualized assessments.
• Accurate and reliable atrophy measurement is crucial for clinical applications.

Purpose of the Study:

• To develop an improved deep learning pipeline for brain atrophy quantification.
• To enhance automated MRI-based segmentation for better diagnostic markers.
• To overcome limitations of existing automated brain atrophy measurement techniques.

Main Methods:

• A deep learning pipeline utilizing tissue similarity regularization for MRI segmentation.
• Training with unlabeled T1-weighted MRI scan pairs and automated target segmentation using FSL.
• Enforcing tissue similarity during training via a weighted loss term penalizing volume differences.
• End-to-end inference for skull stripping and brain tissue segmentation without image interpolation.

Main Results:

• Tissue similarity regularization reduced quantification error and improved volume consistency in short-interval scans.
• The pipeline demonstrated reduced variability in longitudinal atrophy measures.
• Achieved higher effect sizes (Cohen's d) compared to reference methods and SIENA on MIRIAD and ADNI1 datasets.
• Demonstrated significant improvements in differentiating healthy controls from Alzheimer's disease subjects.

Conclusions:

• The proposed deep learning pipeline offers a more accurate and reliable method for brain atrophy quantification.
• Tissue similarity regularization effectively reduces biases and systematic errors in segmentation.
• This advancement is essential for unlocking brain atrophy as a reliable diagnostic and prognostic marker for neurodegenerative diseases.