This exercise will provide a baseline for detecting anomalies in brain MRI images. Makes uses of Scikit-learn's Isolation Forest capabilities to detect outliers. Also, one-class SVM for comparison. Note: dataset removed due to privacy.

• Currently only works for 'grid' images. Extending to full-sized images.
• This exercise will also be repeated with intensity-normalised scans and results compared.
• Further, project test-brains onto 'healthy space' to highlight anomalies (rather than external library tools)

Results:

6 top eigebrains:

outlier detection:

sample of highlighted outlier images: