NeuroBooster: a domain-informed self-supervised learning paradigm tailored for brain MRI analysis.

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Data Analysis

Background:

• Self-supervised learning (SSL) offers an alternative to supervised learning (SL) by reducing the need for extensive labeled data.
• Existing SSL methods are often domain-agnostic, limiting their effectiveness in specialized fields like medical imaging.
• Tailoring SSL paradigms for the medical domain has shown potential for improved performance.

Purpose of the Study:

• To introduce NeuroBooster, a specialized SSL paradigm for brain Magnetic Resonance Imaging (MRI) analysis.
• To leverage automated feature extraction for anatomical features within a regression pretext task.
• To evaluate NeuroBooster's performance in age prediction and Alzheimer's disease classification tasks.

Main Methods:

• NeuroBooster utilizes automated tools to extract anatomical features from brain MRIs.
• These features serve as targets in a regression pretext task for pre-training.
• Experiments involved age prediction (regression) and Alzheimer's vs. healthy subject classification, with 30 randomized seeds for reproducibility.

Main Results:

• NeuroBooster demonstrated superior performance across various brain MRI analysis scenarios.
• In age prediction with only 1% labeled data, NeuroBooster reduced mean absolute error by an average of 5.67 years compared to SL.
• The paradigm showed significant efficacy in both regression and classification tasks.

Conclusions:

• NeuroBooster is a highly effective SSL paradigm specifically designed for brain MRI analysis.
• It significantly reduces the requirement for labeled data, outperforming traditional SL methods.
• This work presents a promising direction for developing domain-specific SSL approaches in medical imaging.