New MRI workflow balances privacy and anatomy with configurable brain extraction

A new paper from Rayeef Ali Khan and Komal Raj Mahantesh introduces a configurable privacy-preserving workflow for structural MRI processing. The method builds on SynthStrip, a deep learning-based brain extraction tool, but instead of producing a single fixed output, it extends the extracted brain mask through morphological expansion. This creates multiple "shell-based" preservation levels—letting researchers choose how much facial and cranial anatomy to strip away. That flexibility is critical because different applications demand different trade-offs: a clinical study might need more anatomical context, while a shared research dataset may require stricter facial feature removal to protect patient identity.

The workflow, implemented in Python using open-source neuroimaging libraries, runs inside the Renku reproducible research environment and includes two key components. An Interactive Preservation Framework lets users visually compare preservation configurations and pick the right output, while an integrated Quality Control Framework provides multi-plane visualization and brain-mask overlay verification to confirm that extraction and defacing are correct. Tested on the publicly available IXI dataset, the system produced anatomically plausible brain extractions with verifiable preservation levels. The main contribution is a modular, reproducible pipeline that gives researchers granular control over privacy versus anatomical fidelity—a practical step toward privacy-oriented neuroimaging and collaborative medical image analysis.