Assessing the Accuracy and Reproducibility of PARIETAL: A Deep Learning Brain Extraction Algorithm.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Manual brain extraction from MRI is time-consuming and inconsistent.
• Automated methods, including deep learning, struggle with variability across scanner vendors and protocols.
• Developing robust automated brain extraction is crucial for clinical research.

Purpose of the Study:

• To introduce and evaluate PARIETAL, a pre-trained deep learning model for brain extraction.
• To assess PARIETAL's reproducibility using scan/rescan analysis.
• To evaluate PARIETAL's robustness across different MRI scanner manufacturers.

Main Methods:

• Retrospective analysis of T1-weighted MRI scans from 21 subjects.
• Acquisition across three different MRI scanners (Siemens, GE, Philips) with varying field strengths and sequences.
• Intracranial cavity volumes were analyzed and compared using parametric permutation tests.

Main Results:

• PARIETAL showed low mean absolute intracranial volume differences in scan/rescan analyses (1.88-4.71 mL).
• It achieved Rank 1 performance on Siemens and GE scanners and Rank 2 on Philips scanners.
• PARIETAL demonstrated the most similar volumetric results between scanners, outperforming other methods.

Conclusions:

• PARIETAL accurately segments the brain and generalizes across different imaging sites.
• No additional training or fine-tuning is required for new data.
• PARIETAL is a publicly available tool for robust automated brain extraction.