From Baby Connectome to White Matter Hyperintensity in Stroke Patients

Date

September 30, 201609/30/2016 3:00pm 09/30/2016 3:00pm From Baby Connectome to White Matter Hyperintensity in Stroke Patients

Brain imaging plays an important part in clinical assessment of brain health across the human life-span. However, the applied methods rely heavily on pre-processing steps, such as definition of brain regions (ROIs) for network analysis or outlining of diseased tissue. In adults, ROIs commonly represent brain areas that are assumed to be functionally coherent. During early development, however, a complete set and locations of ROIs in the brain is yet to be established. This motivates the use of random parcellation schemes with varying numbers of regions or scales. Yet, network measures can be scale dependent, making comparisons across multiple scales challenging and hindering group comparisons. Importantly, the nature of this scale dependence varies between cohorts and can be used in multi-scale frameworks for group comparison. We show that it is possible to differentiate groups based on their scale dependence by applying the framework to a longitudinal preterm cohort, as well as a control group.

In terms of outlining diseased tissue, uncertainties cannot be utilized as such, making fully automated processing pipelines a necessary step to eliminate user-dependent variations. Most pipelines have been developed for research quality scans and subsequently fail in case of the clinical MR data. Thus, it is important to develop dedicated pipelines, which can be applied to images of various resolutions. Recently, we proposed a fully automated, high-throughput white matter hyperintensity (WMH) segmentation pipeline using machine-learning algorithms for WMH detection in clinical stroke data collected from 12 different sites. We show that the outlines generated by these pipelines are in excellent agreement with manually defined regions and apply the pipeline to determine the WMH volume in 2703 stroke patients. Through the development of such a pipeline, which aims to bridge the gap between clinical and research quality scans, we pave the way for scientific discovery by tapping into rich clinical data sets, without the labor intense manual workload.

881 America/Los_Angeles public

Type

Lecture

Time Duration

3:00-4:00 pm

Location

Mission Bay Campus Genentech Hall N-114 Wilsey Seminar Room 1st Floor Lobby

Brain imaging plays an important part in clinical assessment of brain health across the human life-span. However, the applied methods rely heavily on pre-processing steps, such as definition of brain regions (ROIs) for network analysis or outlining of diseased tissue. In adults, ROIs commonly represent brain areas that are assumed to be functionally coherent. During early development, however, a complete set and locations of ROIs in the brain is yet to be established. This motivates the use of random parcellation schemes with varying numbers of regions or scales. Yet, network measures can be scale dependent, making comparisons across multiple scales challenging and hindering group comparisons. Importantly, the nature of this scale dependence varies between cohorts and can be used in multi-scale frameworks for group comparison. We show that it is possible to differentiate groups based on their scale dependence by applying the framework to a longitudinal preterm cohort, as well as a control group.

In terms of outlining diseased tissue, uncertainties cannot be utilized as such, making fully automated processing pipelines a necessary step to eliminate user-dependent variations. Most pipelines have been developed for research quality scans and subsequently fail in case of the clinical MR data. Thus, it is important to develop dedicated pipelines, which can be applied to images of various resolutions. Recently, we proposed a fully automated, high-throughput white matter hyperintensity (WMH) segmentation pipeline using machine-learning algorithms for WMH detection in clinical stroke data collected from 12 different sites. We show that the outlines generated by these pipelines are in excellent agreement with manually defined regions and apply the pipeline to determine the WMH volume in 2703 stroke patients. Through the development of such a pipeline, which aims to bridge the gap between clinical and research quality scans, we pave the way for scientific discovery by tapping into rich clinical data sets, without the labor intense manual workload.

Speakers

Markus Schirmer, PhD
Research fellow at the J. Philip Kistler Stroke Research Centre
Harvard Medical School, Boston, Mass.

Markus Schirmer, PhD is a research fellow at the J. Philip Kistler Stroke Research Centre, part of the Massachusetts General Hospital and Harvard Medical School. Markus did his PhD work at King's College London with a focus on brain connectivity. Now, he also works on stroke, genetics and MRI, addressing the challenges that arise when conducting research with large-scale clinical data.