Related Experiment Video
Updated: Jan 5, 2026
11:03
High-resolution In Vivo Manual Segmentation Protocol for Human Hippocampal Subfields Using 3T Magnetic Resonance Imaging
Published on: November 10, 2015
9.9K
Hippocampal segmentation for brains with extensive atrophy using three-dimensional convolutional neural networks.
Maged Goubran1,2, Emmanuel Edward Ntiri1,2, Hassan Akhavein1,2
1LC Campbell Cognitive Neurology Unit, Hurvitz Brain Sciences Research Program, Sunnybrook Research Institute, University of Toronto, Toronto, Ontario, Canada.
Human Brain Mapping
|October 15, 2019
Summary
A new algorithm, HippMapp3r, accurately and rapidly segments the hippocampus, a key brain region for aging and dementia research. This automated method overcomes limitations of existing tools, offering improved performance and speed for large-scale studies.
Area of Science:
- Neuroimaging and computational neuroscience
- Medical image analysis and artificial intelligence
Background:
- Hippocampal volumetry is crucial for understanding aging and dementia.
- Current automated segmentation methods face challenges with accuracy, availability, and computational cost, especially in cases of brain atrophy or disease.
Purpose of the Study:
- To develop and validate a novel, publicly available 3D convolutional neural network algorithm for automated hippocampal segmentation.
- To address limitations of existing methods, particularly in challenging clinical populations and under adverse imaging conditions.
Main Methods:
- Trained a 3D convolutional neural network (HippMapp3r) on 259 manually segmented hippocampi from elderly individuals with significant brain atrophy and lesions.
- Validated HippMapp3r against four state-of-the-art techniques (HippoDeep, FreeSurfer, SBHV, volBrain, FIRST) using Dice score and correlation coefficient.
- Assessed performance on frontotemporal dementia and vascular cognitive impairment populations and tested robustness on simulated low-quality scans.
Main Results:
- HippMapp3r achieved superior performance, with an average Dice of 0.89 and a correlation coefficient of 0.95, outperforming all compared methods.
- The algorithm demonstrated significantly faster processing speeds, being two orders of magnitude quicker than some existing techniques.
- HippMapp3r exhibited a low outlier rate in disease populations and robustness against corrupted, low-quality scans.
Conclusions:
- HippMapp3r offers a highly accurate, rapid, and robust solution for automated hippocampal segmentation.
- The publicly available pipeline and models facilitate large-scale, multisite research into aging, dementia, and cognitive performance.
- This tool has the potential to significantly advance clinical and research applications of hippocampal volumetry.
