HippMapp3r

HippMapp3r performs precise hippocampal segmentation using three-dimensional convolutional neural networks to enable accurate hippocampal volumetry as a biomarker for aging and dementia.

Key Features:

• Robust segmentation in atrophic brains: Segments hippocampi in brains with significant atrophy, cerebrovascular disease, and lesions that typically challenge conventional algorithms.
• Training on diverse data: The model was trained on 259 bilateral manually delineated segmentations from three studies acquired across multiple sites with various scanners and protocols.
• Superior performance metrics: In validation against HippoDeep, FreeSurfer, SBHV, volBrain, and FIRST, HippMapp3r achieved an average Dice coefficient of 0.89 and a correlation coefficient of 0.95.
• Efficiency: The algorithm is approximately two orders of magnitude faster than many existing methods.
• Low outlier rate: Validation on subjects with frontotemporal dementia and vascular cognitive impairment demonstrated a low outlier rate.
• Robustness to adversarial cases: Tested on real and simulated clinical adversarial cases to evaluate performance on corrupt or low-quality scans.

Scientific Applications:

• Hippocampal volumetry for aging and dementia: Enables quantitative hippocampal volume measurements used as biomarkers in studies of aging and dementia.
• Multisite neuroimaging studies: Provides consistent segmentation across multiple sites, scanners, and protocols for large-scale studies.
• Disease-specific analyses: Supports analyses in frontotemporal dementia and vascular cognitive impairment by providing reliable segmentation in these populations.

Methodology:

The approach trains a three-dimensional convolutional neural network on manually delineated hippocampal segmentations (259 bilateral labels from three studies acquired across multiple sites and scanners), evaluates performance by comparison to HippoDeep, FreeSurfer, SBHV, volBrain, and FIRST with Dice and correlation coefficients, and assesses robustness on subjects with frontotemporal dementia, vascular cognitive impairment, and real and simulated clinical adversarial cases.