Final Report Summary - ABRIM (Advanced Brain Imaging with MRI)
Specific Objectives
In this project we bring for the first time modern methods of cognitive neuroimaging using MRI into routine application. This will bring significant benefits both for the clinic and for neuroscience. By exploiting recent dramatic increases in scanning speed we will develop a protocol that performs an assessment of brain connectivity plus a high spatial resolution evaluation of the brain within half an hour.
The measurements described above have hitherto been too time consuming for application in a single imaging session, or the session has been too long to be tolerated by patients and the elderly. Furthermore the application of sophisticated graph theoretical analyses has remained in the domain of neuroimaging research laboratories, and has not been developed for application at the single subject level. In this project we will achieve the following:
1. Design and implement a measurement protocol capable of obtaining measures of functional and anatomical connectivity, grey matter volume, spontaneous BOLD activity, susceptibility weighted imaging, and myelinisation degree within a time frame acceptable to patients, and of the order of 30 minutes.
2. Perform a connectivity analysis on the basis of a standard anatomical atlas, and derive maps of node degree and betweenness. Visualise the outcomes of the other measurements alongside these.
3. Collect data for a data-base of healthy subjects stratified on the basis of age and gender. This will be used for neuroscientific as well as clinical purposes, and in particular to identify the normal range of values for all parameters. By distributing these values rather than the whole data base it will be possible to determine whether an individual deviates significantly from the age and gender related norm.
Work performed in period September 2015-August 2017
Daniel Gomez has continued to work on fast acquisition schemes for fMRI. In addition to the multiband and multi-echo work that he performed in the first period he has now also implemented an even faster technique called MESH (multiband echo-shifted EPI). The choice of implementing this sequence was made on the basis of results obtained in the first reporting period, that showed that it seemed to be better to acquire data as fast as possible rather than the slower multi-echo approach. The whole package of work that he has developed was released by Siemens as a works in progress package in Autumn 2016. This included multiband imaging, multiband multi-echo and MESH. He has subsequently been working on a major project to determine whether multiband (MB) or multiband multi-echo (MBME) is superior for both resting state fMRI acquisitions, and for task based fMRI. This has involved collecting data with both sequences in repeat sessions with the same subjects. Extensive data cleaning has been performed with different post-processing schemes to remove the effects of physiological noise.
Although Zahrah Shams contract was terminated during the first reporting period she has continued to work on determining the ideal method to estimate myelin content from MRI protocols. This has resulted in a manuscript that is currently under revision. In this work she has compared two commonly used techniques: MP2RAGE which obtains all necessary data from a single acquisition to produce maps of R1 (R1=1/T1, and is believed to be directly related to myelin concentration), and the ratio of two T1/T2 weighted images. The total duration of each measurement was kept constant. The comparison should provide important information for the neuroimaging community as to which of these is the best.
Kwok-shing Chan was the replacement for Zahrah Shams, and has worked on the further development of the STIFT (structure tensor informed fibre-tracking) method at 3T. He has improved the data acquisition protocol, and assessed whether the structure tensor is better calculated on the basis of T2*-maps, or quantitative susceptibility maps including diamagnetic susceptibility map-weighted images (dSMWI). During his period at Siemens he worked on a processing and display pipeline to process T1-weighted images and multi-echo GRE images. The T1w processing pipeline consists of brain extraction and tissue segmentation methods; while the mGRE pipeline provides brain extraction, SWI venography and T2* mapping methods.
We have developed a comprehensive acquisition protocol for myelin mapping, diffusion-weighted and susceptibility weighted imaging, resting and task-based fMRI. Acquisition has started on a cohort of 300 subjects spanning the human lifespan.
Main Results
In the comparison between MB and MBME when examining the resting state networks of the brain, the two methods showed similar sensitivities (statistical Z-scores) when the data were not cleaned. The commonly used multi-echo ICA (MEICA) method was shown to be inferior for cleaning data to FIX method, which has generally only been used on MB and not on MBME data. MEICA is considered problematic by us as it removes too much of the variance from the data. There seems to be little difference between MB and MBME if they are both cleaned optimally using FIX, but MBME shows better performance in regions of the brain having poorer homogeneity of the static field.
In the comparison of cortical mapping methods both show similar and highly replicable cortical maps enhancing regions traditionally found as being highly myelinated. T1w/T2w maps have shown to be more robust to motion related artifacts, which was attributed to relying on two separate shorter acquisitions rather than one longer one. Both methods showed similar reproducibility in test re-test experiments within the same subject, but R1 maps have shown increased differentiation across neighbouring cortical regions and across subjects. This could make R1 a preferable metric in longitudinal studies while T1w/T2w maps showed a higher reproducibility of the average myelination across subjects for different Brodmann areas which could make it a preferable metric for cortical segmentation.
We have demonstrated that it is possible to perform STIFT using data acquired only at 3T. In the original study DWI and GRE data were collected at 3T and 7T respectively. In addition to the magnitude of GRE data, the phase of GRE also possesses strong white matter contrast that can be used for STIFT. We could reproduce the results of the study at 7T to show improved differentiation of some white matter tracts with STIFT. However, the total connectivity matrix was not improved by using STIFT irrespective of the way in which the susceptibility tensor maps were generated.
Expected Final Results
• There will be a conclusion as to whether MB or MBME is preferred for both resting state and fMRI acquisitions.
• T1w/T2w will probably be the method of choice for assessing myelin in vivo.
• The STIFT technique offers only a marginal improvement for fibre-tracking with DWI.
• There will be a 300 subject database showing the development of connectivity with age.
Contact
Website: https://www.ru.nl/donders/collaborations/projects/abrim/;
Email: david.norris@donders.ru.nl
In this project we bring for the first time modern methods of cognitive neuroimaging using MRI into routine application. This will bring significant benefits both for the clinic and for neuroscience. By exploiting recent dramatic increases in scanning speed we will develop a protocol that performs an assessment of brain connectivity plus a high spatial resolution evaluation of the brain within half an hour.
The measurements described above have hitherto been too time consuming for application in a single imaging session, or the session has been too long to be tolerated by patients and the elderly. Furthermore the application of sophisticated graph theoretical analyses has remained in the domain of neuroimaging research laboratories, and has not been developed for application at the single subject level. In this project we will achieve the following:
1. Design and implement a measurement protocol capable of obtaining measures of functional and anatomical connectivity, grey matter volume, spontaneous BOLD activity, susceptibility weighted imaging, and myelinisation degree within a time frame acceptable to patients, and of the order of 30 minutes.
2. Perform a connectivity analysis on the basis of a standard anatomical atlas, and derive maps of node degree and betweenness. Visualise the outcomes of the other measurements alongside these.
3. Collect data for a data-base of healthy subjects stratified on the basis of age and gender. This will be used for neuroscientific as well as clinical purposes, and in particular to identify the normal range of values for all parameters. By distributing these values rather than the whole data base it will be possible to determine whether an individual deviates significantly from the age and gender related norm.
Work performed in period September 2015-August 2017
Daniel Gomez has continued to work on fast acquisition schemes for fMRI. In addition to the multiband and multi-echo work that he performed in the first period he has now also implemented an even faster technique called MESH (multiband echo-shifted EPI). The choice of implementing this sequence was made on the basis of results obtained in the first reporting period, that showed that it seemed to be better to acquire data as fast as possible rather than the slower multi-echo approach. The whole package of work that he has developed was released by Siemens as a works in progress package in Autumn 2016. This included multiband imaging, multiband multi-echo and MESH. He has subsequently been working on a major project to determine whether multiband (MB) or multiband multi-echo (MBME) is superior for both resting state fMRI acquisitions, and for task based fMRI. This has involved collecting data with both sequences in repeat sessions with the same subjects. Extensive data cleaning has been performed with different post-processing schemes to remove the effects of physiological noise.
Although Zahrah Shams contract was terminated during the first reporting period she has continued to work on determining the ideal method to estimate myelin content from MRI protocols. This has resulted in a manuscript that is currently under revision. In this work she has compared two commonly used techniques: MP2RAGE which obtains all necessary data from a single acquisition to produce maps of R1 (R1=1/T1, and is believed to be directly related to myelin concentration), and the ratio of two T1/T2 weighted images. The total duration of each measurement was kept constant. The comparison should provide important information for the neuroimaging community as to which of these is the best.
Kwok-shing Chan was the replacement for Zahrah Shams, and has worked on the further development of the STIFT (structure tensor informed fibre-tracking) method at 3T. He has improved the data acquisition protocol, and assessed whether the structure tensor is better calculated on the basis of T2*-maps, or quantitative susceptibility maps including diamagnetic susceptibility map-weighted images (dSMWI). During his period at Siemens he worked on a processing and display pipeline to process T1-weighted images and multi-echo GRE images. The T1w processing pipeline consists of brain extraction and tissue segmentation methods; while the mGRE pipeline provides brain extraction, SWI venography and T2* mapping methods.
We have developed a comprehensive acquisition protocol for myelin mapping, diffusion-weighted and susceptibility weighted imaging, resting and task-based fMRI. Acquisition has started on a cohort of 300 subjects spanning the human lifespan.
Main Results
In the comparison between MB and MBME when examining the resting state networks of the brain, the two methods showed similar sensitivities (statistical Z-scores) when the data were not cleaned. The commonly used multi-echo ICA (MEICA) method was shown to be inferior for cleaning data to FIX method, which has generally only been used on MB and not on MBME data. MEICA is considered problematic by us as it removes too much of the variance from the data. There seems to be little difference between MB and MBME if they are both cleaned optimally using FIX, but MBME shows better performance in regions of the brain having poorer homogeneity of the static field.
In the comparison of cortical mapping methods both show similar and highly replicable cortical maps enhancing regions traditionally found as being highly myelinated. T1w/T2w maps have shown to be more robust to motion related artifacts, which was attributed to relying on two separate shorter acquisitions rather than one longer one. Both methods showed similar reproducibility in test re-test experiments within the same subject, but R1 maps have shown increased differentiation across neighbouring cortical regions and across subjects. This could make R1 a preferable metric in longitudinal studies while T1w/T2w maps showed a higher reproducibility of the average myelination across subjects for different Brodmann areas which could make it a preferable metric for cortical segmentation.
We have demonstrated that it is possible to perform STIFT using data acquired only at 3T. In the original study DWI and GRE data were collected at 3T and 7T respectively. In addition to the magnitude of GRE data, the phase of GRE also possesses strong white matter contrast that can be used for STIFT. We could reproduce the results of the study at 7T to show improved differentiation of some white matter tracts with STIFT. However, the total connectivity matrix was not improved by using STIFT irrespective of the way in which the susceptibility tensor maps were generated.
Expected Final Results
• There will be a conclusion as to whether MB or MBME is preferred for both resting state and fMRI acquisitions.
• T1w/T2w will probably be the method of choice for assessing myelin in vivo.
• The STIFT technique offers only a marginal improvement for fibre-tracking with DWI.
• There will be a 300 subject database showing the development of connectivity with age.
Contact
Website: https://www.ru.nl/donders/collaborations/projects/abrim/;
Email: david.norris@donders.ru.nl