Machine Learning for Brain MRI Data Harmonisation: A Systematic Review.

Area of Science:

• Medical Imaging
• Data Science
• Neuroscience

Background:

• Multi-center Magnetic Resonance Imaging (MRI) data exhibits heterogeneity due to scanner and site variations.
• Data harmonization is crucial to mitigate these variations for reliable analysis.
• Machine learning (ML) has emerged as a promising tool for addressing MRI data heterogeneity.

Purpose of the Study:

• To systematically review and evaluate the performance of various ML algorithms in harmonizing MRI data.
• To summarize findings on implicit and explicit MRI data harmonization techniques.
• To provide guidelines for current methods and identify future research directions in ML-based MRI harmonization.

Main Methods:

• A comprehensive literature search was conducted across PubMed, Web of Science, and IEEE databases up to June 2022.
• Systematic review methodology adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines was employed.
• Quality assessment of included publications was performed using derived criteria.

Main Results:

• Analysis of 41 articles (2015-2022) revealed both implicit (n=21) and explicit (n=20) ML-based MRI harmonization approaches.
• Structural MRI (n=28) was the most common modality, followed by diffusion MRI (n=7) and functional MRI (n=6).
• Various ML techniques have been applied across different MRI modalities.

Conclusions:

• ML techniques are effective for harmonizing diverse MRI data types.
• A lack of standardized evaluation metrics across studies hinders consistent assessment.
• Future research should focus on developing consistent evaluation methods; ML-harmonized data shows promise for downstream tasks but requires cautious interpretation.