Related Experiment Video
Updated: Aug 31, 2025
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
Goal-specific brain MRI harmonization
Lijun An1, Jianzhong Chen1, Pansheng Chen1
1Centre for Sleep and Cognition (CSC) & Centre for Translational Magnetic Resonance Research (TMR), Yong Loo Lin School of Medicine, National University of Singapore, Singapore; Department of Electrical and Computer Engineering, National University of Singapore, Singapore; N.1 Institute for Health and Institute for Digital Medicine (WisDM), National University of Singapore, Singapore.
Abstract:
There is significant interest in pooling magnetic resonance image (MRI) data from multiple datasets to enable mega-analysis. Harmonization is typically performed to reduce heterogeneity when pooling MRI data across datasets. Most MRI harmonization algorithms do not explicitly consider downstream application performance during harmonization. However, the choice of downstream application might influence what might be considered as study-specific confounds. Therefore, ignoring downstream applications during harmonization might potentially limit downstream performance. Here we propose a goal-specific harmonization framework that utilizes downstream application performance to regularize the harmonization procedure. Our framework can be integrated with a wide variety of harmonization models based on deep neural networks, such as the recently proposed conditional variational autoencoder (cVAE) harmonization model. Three datasets from three different continents with a total of 2787 participants and 10,085 anatomical T1 scans were used for evaluation. We found that cVAE removed more dataset differences than the widely used ComBat model, but at the expense of removing desirable biological information as measured by downstream prediction of mini mental state examination (MMSE) scores and clinical diagnoses. On the other hand, our goal-specific cVAE (gcVAE) was able to remove as much dataset differences as cVAE, while improving downstream cross-sectional prediction of MMSE scores and clinical diagnoses.
Insights
Goal-specific harmonization improves magnetic resonance image (MRI) data pooling. This new framework enhances downstream task performance by regularizing harmonization, outperforming standard methods in predicting clinical outcomes.
Area of Science:
- Neuroimaging
- Medical Data Science
- Machine Learning in Healthcare
Background:
- Pooling magnetic resonance image (MRI) data across multiple datasets is crucial for large-scale studies (mega-analysis).
- Data heterogeneity between sites necessitates harmonization, but current methods often overlook downstream application performance.
- Ignoring downstream task relevance may limit the effectiveness of MRI harmonization techniques.
Purpose of the Study:
- To introduce a goal-specific harmonization framework that leverages downstream application performance to guide the harmonization process.
- To integrate this goal-specific approach with deep learning-based harmonization models, such as conditional variational autoencoders (cVAE).
- To evaluate the framework's ability to reduce dataset differences while preserving biologically relevant information for clinical predictions.
Main Methods:
- Development of a goal-specific harmonization framework (gcVAE) that incorporates downstream performance metrics into the regularization process.
- Integration of gcVAE with a conditional variational autoencoder (cVAE) model for MRI harmonization.
- Evaluation using three multi-continental MRI datasets (2787 participants, 10,085 scans) and assessment of downstream prediction accuracy for Mini Mental State Examination (MMSE) scores and clinical diagnoses.
Main Results:
- The standard cVAE model reduced dataset differences more effectively than the ComBat model but diminished predictive power for MMSE scores and clinical diagnoses.
- The proposed goal-specific cVAE (gcVAE) achieved comparable reduction in dataset differences to cVAE.
- gcVAE significantly improved the cross-sectional prediction accuracy of MMSE scores and clinical diagnoses compared to both ComBat and standard cVAE.
Conclusions:
- Goal-specific harmonization using gcVAE offers a superior approach for pooling MRI data by balancing data harmonization with the preservation of predictive biological information.
- This framework enhances the utility of multi-site MRI datasets for downstream clinical applications and neuroscientific research.
- Optimizing harmonization based on specific downstream tasks leads to more effective data integration and improved predictive modeling in neuroimaging.
