Skip to content

Open-Source Tool Performs Whole-Brain Segmentation of Longitudinal MRI Scans

Key findings

  • Researchers at Massachusetts General Hospital have developed a new method for whole-brain segmentation of longitudinal MRI scans, with or without simultaneous segmentation of white matter lesions
  • The tool is generalizable across different scanner platforms, field strengths, acquisition protocols, and image resolutions without any retraining or retuning
  • Validation conducted using 4,553 brain MRI scans showed the new tool has better test–retest reliability than benchmark methods and is more sensitive to disease-related changes in Alzheimer's disease and multiple sclerosis
  • The tool is publicly available as part of the FreeSurfer open-source software package for analysis of neuroimaging data

Over the past two decades, longitudinal MRI scanning has provided valuable insights into healthy brain development, the effects of aging on the brain, and the pathogenesis of neurodegenerative diseases. Furthermore, longitudinal MRI is an important way to monitor individual patients with neurodegenerative diseases and evaluate various treatments in clinical trials.

Many dedicated tools have been developed to analyze longitudinal brain MRI scans. The most comprehensive is the one distributed with FreeSurfer, a suite of software for automated quantification of the functional, connectional and structural properties of the brain. FreeSurfer runs on a wide variety of hardware and software platforms and is open-source.

The FreeSurfer tool for longitudinal segmentation (Aseg-Long) segments many neuroanatomical structures simultaneously, and it readily processes data at more than two time points. However, it's designed for T1-weighted (T1w) scans. It isn't well suited for analyzing white matter lesions or other pathologies that are better visualized using other MRI contrasts, such as T2w or fluid-attenuated inversion recovery.

Researchers at Massachusetts General Hospital have developed a method for automatically segmenting dozens of neuroanatomical structures from longitudinal MRI scans, using a model-based approach that accepts multi-contrast data as input and can segment white matter lesions simultaneously. Stefano Cerri, PhD, research fellow at the Martinos Center for Biomedical Imaging at Mass General, Koen Van Leemput, PhD, an investigator at the Martinos Center, and colleagues report in NeuroImage: Clinical.


The team built on a previously validated tool for whole-brain segmentation of cross-sectional, Sequence Adaptive Multimodal SEGmentation (SAMSEG). SAMSEG segments 41 anatomical structures from brain MRI and can be fully adapted to different MRI scanner platforms and contrasts. The researchers give details in their report on how they extended SAMSEG for longitudinal scans.

Like its predecessor, the new method does not put any constraints on the MRI scanner used or the imaging protocol. Moreover, it accommodates any number or timing of longitudinal follow-up scans.

The tool can also segment white matter lesions from longitudinal scans acquired with conventional MRI sequences. This allows users to simultaneously track lesion evolution and morphological changes in various brain structures, for example in patients with multiple sclerosis.

Extensive Validation

Validation of the new tool was conducted using 4,553 brain MRI scans acquired with different scanners, field strengths, acquisition protocols and image resolutions. The datasets included single- and multi-contrast longitudinal scans with a range of time gaps and total number of time points, from both healthy individuals and patients suffering from Alzheimer's disease or multiple sclerosis.

No retraining or retuning of the tool was needed. It produced more reliable segmentations in scan–rescan settings compared with either Aseg-Long or SAMSEG. It was also more sensitive to differences in longitudinal changes between patient groups.

Expected Advantages

Use of the new tool may reduce the number of subjects needed in longitudinal research studies, result in better patient stratification and allow more precise evaluation of treatment efficacy. It is now publicly available as part of FreeSurfer.

Learn more about the FreeSurfer software suite

Learn more about the Martinos Center for Biomedical Imaging


Jyrki Ahveninen, PhD, and colleagues believe sparse brain activity, where only a few neurons in a population fire at a given time, might contribute to the fingerprint patterns of activity evident in multi-voxel pattern analysis results from functional MRI—and give rise to perceptual, emotional and cognitive functions.


Randy L. Buckner, PhD, and colleagues have provided direct support for a third somatomotor map in the vermis of the cerebellum by using intensive, repeated functional MRI scanning of individuals performing movements across multiple body parts.