Automated segmentation of midbrain structures with high iron content
Benjamín Garzón1, Rouslan Sitnikov2, Lars Bäckman1
1Aging Research Center (ARC), Karolinska Institute and Stockholm University, Sweden.
Neuroimage
|June 13, 2017
Summary
We developed a new automated method to segment midbrain structures like the substantia nigra (SN), subthalamic nucleus (STN), and red nucleus (RN) using quantitative susceptibility mapping (QSM) for improved neuroimaging analysis.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Neuroanatomy
Background:
- Midbrain structures like the substantia nigra (SN), subthalamic nucleus (STN), and red nucleus (RN) are crucial in neuroimaging research.
- Automated segmentation methods are needed to efficiently analyze these iron-rich structures.
Purpose of the Study:
- To present a novel automated segmentation method for the SN, STN, and RN.
- To leverage quantitative susceptibility mapping (QSM) for enhanced segmentation accuracy.
Main Methods:
- Developed an algorithm using spatial priors derived from non-linear registration of training labels.
- Employed a Gaussian mixture model with smoothness constraints for intensity-based segmentation.
- Validated the method on manual segmentations from 40 healthy subjects.
Main Results:
- Achieved high average Dice scores: 0.81 for SN, 0.66 for STN, and 0.88 for RN.
- Demonstrated significant correlations between manual and automated segmentation volumes.
- Observed lower accuracy on R2* and FLAIR images compared to QSM.
Conclusions:
- The novel QSM-based method provides accurate automated segmentation of key midbrain structures.
- The automated segmentations are comparable to manual ones for detecting age-related iron content differences.
