NA-MIC Project Weeks

Back to Projects List

Improving QC protocols for AMP SCZ Clinical and MRI Data

Key Investigators

• Sylvain Bouix (école de technologie supérieure, Canada)
• Owen Borders (Psychiatry Neuroimaging Lab, U.S.)
• Keerthana Srinivasan (Psychiatry Neuroimaging Lab, U.S.)

Project Description

We will be adding anomaly detection algorithms to clinical data and MRI data in the AMP SCZ project.

Objective

1. We would like to achieve MRI QC reports that accurately correlate to human QC ratings, in addition to uncovering new QC rules for clinical forms using machine learning anomaly detection algorithms.

Approach and Plan

1. For the clinical data, we will use DBSCAN to search for correlations between every variable and flag cases that deviate the most.
2. For the MRI data, we will artificially add artifacts to clean MRI scans and train a neural network to rank the severity.

Progress and Next Steps

1. Scanned the database for site distribution outliers, string anomalies, and time series anomalies that allowed several new quality control rules to be developed.
2. Created a web interface for MRI data that allowed various synthetic anatomical and scanner features to be visualized.
3. Used the synthetic anatomical and scanner augmentations to train a robust MRI QC model.
4. Next steps involve calculating the correlation between the automatic MRI QC and scores by human raters.

Illustrations

No response

Background and References

No response

Funding

This work is supported by NIH U24MH137171.