Final Activity Report Summary - LONGSERIMULTIQMRIVOL (Long-term Serial Multi-contrast Quantitative MR Imaging and Volumetry)
This project focused on the development of software to identify and measure two small, but important, brain structures, called the hippocampi and the amygdale, from magnetic resonance imaging (MRI) scans. In particular, it is important to measure changes in these structures in time in patients with chronic neurological conditions such as epilepsy and dementia. The detection and measurement process is called image segmentation. Image segmentation can be made through a time-consuming and subjective process called manual segmentation. Therefore, the overall objectives of on-going software developments are to increase the measurement's reliability and reduce the need for human intervention (automation).
The main aim of this project was to improve our ability to detect and quantify changes in the hippocampi and amygdale across time. The project comprised three main objectives, namely:
1) use of multi-contrast information (MRI scans acquired with different parameters showing different contrast characteristics) to improve the automatic segmentation of the hippocampus;
2) improving and adapting the automatic segmentation of the hippocampus to make it more sensitive to detection of changes in longitudinal studies together with a comparison with probabilistic atlases methods; and
3) comparison of data acquired on MRI scanners with magnetic fields of 1.5Tesla and 3Tesla, to allow correcting the differences in volumes of the hippocampus in long term longitudinal studies involving several scanners.
The project built on the result of the fellow's previous PhD work, software which required adaptation and improvements for use with data acquired on a different MRI scanner. This software offered semi-automated segmentation of the hippocampi and amygdale, requiring expert user intervention, which can make the method unreliable. In the course of the initial work, it was realised that probabilistic atlases, which are essentially statistical images of brain structures derived from labour-intensive image analysis, offered the possibility of removing the need for any user intervention, to render the software fully automatic.
By modifying the software to take into account prior knowledge in the form of a probabilistic atlas we have recently been able to validate a fully automatic and accurate segmentation of the hippocampi and amygdale. This new development has potential important implications for the way in which numerous scientific investigations are performed.
In addition, a new scanning protocol was designed to study ways to improve sensitivity in studies of change in the two brain structures of interest across time (longitudinal change). The results have yet to be fully analysed but we hope to be able to demonstrate unprecedented degrees of sensitivity to change in data acquired using the same MRI scanner and using different scanners. The latter makes segmentation more difficult because MRI scanners cannot be calibrated in way that makes images and the measurements derived from them easily comparable.
By scanning a number of subjects twice at 1.5T and twice at 3T we are able to evaluate those differences and to calculate correction factors necessary that facilitate comparison and therefore the sensitivity to change.
The main aim of this project was to improve our ability to detect and quantify changes in the hippocampi and amygdale across time. The project comprised three main objectives, namely:
1) use of multi-contrast information (MRI scans acquired with different parameters showing different contrast characteristics) to improve the automatic segmentation of the hippocampus;
2) improving and adapting the automatic segmentation of the hippocampus to make it more sensitive to detection of changes in longitudinal studies together with a comparison with probabilistic atlases methods; and
3) comparison of data acquired on MRI scanners with magnetic fields of 1.5Tesla and 3Tesla, to allow correcting the differences in volumes of the hippocampus in long term longitudinal studies involving several scanners.
The project built on the result of the fellow's previous PhD work, software which required adaptation and improvements for use with data acquired on a different MRI scanner. This software offered semi-automated segmentation of the hippocampi and amygdale, requiring expert user intervention, which can make the method unreliable. In the course of the initial work, it was realised that probabilistic atlases, which are essentially statistical images of brain structures derived from labour-intensive image analysis, offered the possibility of removing the need for any user intervention, to render the software fully automatic.
By modifying the software to take into account prior knowledge in the form of a probabilistic atlas we have recently been able to validate a fully automatic and accurate segmentation of the hippocampi and amygdale. This new development has potential important implications for the way in which numerous scientific investigations are performed.
In addition, a new scanning protocol was designed to study ways to improve sensitivity in studies of change in the two brain structures of interest across time (longitudinal change). The results have yet to be fully analysed but we hope to be able to demonstrate unprecedented degrees of sensitivity to change in data acquired using the same MRI scanner and using different scanners. The latter makes segmentation more difficult because MRI scanners cannot be calibrated in way that makes images and the measurements derived from them easily comparable.
By scanning a number of subjects twice at 1.5T and twice at 3T we are able to evaluate those differences and to calculate correction factors necessary that facilitate comparison and therefore the sensitivity to change.