Sunrise Educational Session: Controversies in Diffusion & Functional MRI


Wednesday 11May 2016

Ultra-fast fMRI of task and resting state

Author: Vesa Kiviniemi vesa.kiviniemi@oulu.fi

Highlights

Comparison of task vs. resting state research with a pre-surgical flavour

Learn about ultra-fast, critically sampled fMRI data

Improve your knowledge on baseline brain physiology and task causality measurements from critically sampled data

TARGET AUDIENCE : neuroscientists, PhD-students, clinical/research personel doing (pre-surgical) fMRI

OBJECTIVES: Understand the intertwined nature of resting state vs. task based fMRI scanning. Acknowledge the basic principles and the importance of correct statistical approaches to analysis of both datasets. Learn pro’s and con’s of both approaches with a special emphasis on pre-surgical planning. Get a feel of what might be missed currently and learn ways to improve accuracy in both approaches via ultra-fast scanning (in the near future).

PURPOSE : Open a historical perspective on task vs. resting state analysis of brain function. Show methodological issues regarding analysis of brain functionality in both rest and task. Shed light into future prospects of ultra-fast functional brain imaging.

Historical background Task based fMRI has become the leading functional neuroimaging method over the past quarter of a century since it’s original discoveries in early 1990’s by Ogawa and others. The classical task-activation approach has been to make the scanned person do intermitted, cued tasks between “resting” state, i.e. between period of no tasks. From early 1990’s till roughly 2005 the task studies were dominating the field almost exclusively.

Despite the dominance of the task studies 1995 Bharat Biswal noticed with Jim Hyde tha t spontaneous brain signal does have a hidden message in it; the seemingly spontaneous fluctuations of the detected fMRI signal are highly synchronous in functionally connected brain regions. For ten years this discovery was only studied by a few pioneering scientists that thought differently. The task studies are based on the BOLD signal increase, where the T2* weighted signal increases 3-5 seconds after the onset of the stimulus. This signal increase is most often compared to the resting state before and after the stimulus. Recent theories on the origins of resting state signal fluctuations suggest that repeating avalanches of neuronal activity induce co-activation patterns that become detected as repeated BOLD activation patterns.

As the brain actually never rests, the observed signal changes can be markedly affected by these baseline signal activations during the “resting state”. As shown beautifully by Fox et al in 2005, the spontaneous motor cortex fluctuations ipsi-lateral to the finger tapping explain 60 % of the signal noise in the contra-lateral motor cortex that was producing the finger tapping commands. This along with the detection of rest-activated default mode network markedly increased the interest in the resting state fluctuations. Later it has been shown that the task activity signal as well as measured performance of the subject could be predicted from connected network fluctuations of other areas. Furthermore a meta analysis of 1700 task studies in nearly 30 000 subjects revealed the same networks as resting state data from 36 subjects.

In pre-surgical planning the task-activation has gained much interest since one can safely map the locations of specific brain functions that may be jeopardized in the brain surgery. If for example primary functional areas of a given task such as Broca’s area in speech or motor cortex on motion control is in proximity of a operated area (Tumor, AVM,etc) one can obtain a better post-operative result in terms of QALY if one can avoid the removal of a given functional area.

However, in some cases the tumor may be so close to the cortical area that the control of the function is already compromised or the patient co-operation to the task is limited. Furthermore AVM’s may have shunted blood flow preventing hemodynamic responses required for BOLD fMRI. In such cases resting state scanning might offer an alternative. The problematic issue with resting state data is then the lack of specific knowledge in what functions a given connectivity pattern may represent, especially in cases where the anatomical knowledge may not be used to help the identification of the source.Fig.1 illustrates the similarity of results from task and rest analyses.

Analysis tools. Initially the first task studies were performed simply by subtracting the signal increase from the baseline signal, much like in DSA. Later statistical methods were introduced with a more developed hemodynamic response function that describes the common response to task. These tools enabled more robust detection of activity from the resting state fluctuations. Soon publicly available software packages like AFNI, SMP and FSL, to name a few, occurred supported by the internet. It needs to be noted that the fast improvement of functional analysis tools largely depends on the free access of all scientist to these tools without the interference of financial issues.

As for the resting state the first ideas on the origin of the detected fluctuations were based on known physiological factors, such as spontaneous vasomotor waves and such physiological phenomena. Seed voxel time domain signal cross correlation with all brain voxels was the first way to detect the fluctuations. Later frequency domain methods were also introduced followed by what I personally think is the best option for functional brain analysis; the independent component analysis (ICA). ICA offers additional value compared to other methods due to it’s ability to separate noise/motion dominated components from neurofunctional sources in resting state data. Both of these methods were originally used first in task activation studies.

A key philosophical question between the resting vs. task state is the assumed control of neuronal events during scanning. While in task state one knows temporally when to start looking for the targeted activity, in resting state there is no control over any of the measured parameters. In the case on non-deterministic signal the best choice in analyzing the signal is a statistical approach. ICA offers a strong statistical way to analyze both task and resting state signal. ICA can separate statistically independent signal sources based on their joint density distributions.

Recent developments on analyzing neuronal avalanches and physiological pulsations such as QPP analysis by Shella Keilholz's group enable the detection of spatially moving activity patterns. Fig's 2-3 show an example of default mode avalanches and physiological pulsations that energize the brain glymphatic system.

Classic interleaved EPI vs. ultra-fast 3D imaging. It is surprising how much of the brain MRI signal is relevant information. Some of the signal is thermal/scanner noise but it has markedly diminished in recent years thanks to excellent progress in engineering in the field. Nowadays scanner noise it can be considered a minor issue and dealt with.

Classical relatively slowly sampled (TR 2-3 sec) whole brain data fMRi data is not fast enough for measuring physiological pulsations. The pulsations have been considered to be noise that shadows neurophysiological information. But with the recent discovery of brain waste clearance mechanism the glymphatic system the cardiorespiratory pulsations are likely to become of vital clinical relevance.

Ultra-fast fMRI sequences of 3D brain coverage < 100 ms can critically sample the physiological pulsations; the cardiorespiratory pulses do not aliases in the signal as they inevitably do with longer TR’s (> 400 ms). Furthermore, the data has much more statistical power that makes single subject analysis more accurate allowing maybe clinically relevant decision making at individual level. Furthermore recent ultra-fast imaging results show close resemblence to MEG results indicating that causality mapping of spreading neuronal events is possible.

Combining multimodal brain signal measurement tools is one way of finding out exact signal source mechanism that shape the signals we measure. Putative information can nowadays be critically sampled in synchrony with the fMRI/BOLD signal and this produces massive amounts of data. Data mining and exploratory analyses become very important in understanding the interrelationships of these synchronous measures. New areas of glymphatic pulsation imaging and avalanche detection are becoming more realistic. These may enable comprehensive analysis of the global physiological status of the human brain enabling finally clinically valuable information of the brain with fMRI.