So today, most of the fMRI data that we've been talking about has been so called task based fMRI. So this is data that's acquired while the subject performs an explicit task. However, it's clear that the brain is active even if the subject is not performing a task. And in fact, according to certain estimates, task-related changes in neuronal metabolism only account for about 5% of the brain's total energy consumption. So, resting state fMRI is a relatively new approach used to identify bold changes in multiple brain regions while subjects lie in the scanner but don't perform a task. In particular, it has been shown that fluctuations in the low-frequency portion of the BOLD signal show strong correlations across spatially distant regions to the brain. This is thought to be caused by fluctuations in spontaneous neural activation, that exact mechanism remains unclear. So neuroscientists are increasingly interested in studying the correlation between these spontaneous BOLD signal across the brain regions in order to learn about its intrinsic functional connectivity. Because of the lack of task, resting state fMRI is attractive as it removes the burden of experimental design, subject compliance, and training demands. So for these reasons, it's particularly attractive for studies of development and clinical populations, so subjects that might not be able to perform a task. Also it's very easy to add on a resting state scan at the end of your task based experiments. And so for these reasons the availability of resting state data's exploded in the past couple of years. So research has already revealed large-scale spatial patterns of coherent signal in the brain during rest and these correspond to these functionally relevant resting-state networks. These resting-state networks are thought to reflect the neuronal baseline activity of the brain. A number of resting-state networks have been consistently observed both across groups of subjects and in repeated scanning sessions on the same subject. These resting state networks tend to be localized to grey matter and are thought to reflect functional systems supporting core perceptual and cognitive processes. And regions that are co-activated during active tasks also show resting state connectivity. So brain regions with similar functionality tend to express similar patterns of spontaneous BOLD response. Sometimes subsets of these resting state networks appear to be either up or down-regulated during specific cognitive tasks. Here's an example of eight of the most common and consistently observed resting state networks identified using ICA or independent component analysis. And this is from a paper by Colon colleagues, here we see, for example, a network corresponding to the visual cortex, one corresponding to the auditory cortex, another to the somatomotor cortex. And finally, we see one corresponding to default mode network, which is a very commonly occurring network in resting state fMRI. So resting state fMRI as I mentioned previously is based on studying low frequency BOLD fluctuations. So functionally relevant, spontaneous BOLD oscillations that are found in the lower frequency ranges typically vary between 0.01 and 0.08 Hz. This is separable from single frequencies corresponding to respiratory and cardiovascular signal. Typically, resting state experiments are in the order of five or ten minutes long, though the identification of the optimal length of time to perform resting state hasn't been identified yet. And the possible need for multiple sessions remain an open issue. In addition, there's really no consensus over whether data should be collected while subjects are asleep or awake, or with eyes open or eyes closed. The pre-processing of resting state data follows the same general pipeline as applied to standard task-based fMRI. However, there are few important differences. Because, we're interested in certain frequency bands that the data is often band-pass filtered at 0.01 to 0.08 because that is sort of the sweet spot in resting state fMRI. And so the traditional high pass tempering filtering applied to task-based fMRI data is often thought to be overly aggressive with respect to removing some of these relevant frequency information. It's also been shown that non-neuronal physiological signals may interfere with the resting state BOLD data. So removal of confounding signals, such as the respiratory or cardiovascular noise, will considerably improve the quality of the data which can be attributed to neural activation. And it's therefore become common practice in resting state fMRI to monitor such signals and retrospectively try to correct for them in post-acquisition. In addition, usually the global mean signal, at least six motion parameters, the CSF and white matter signals are also commonly removed as nuisance variables, to reduce the effects of head motion and non-neuronal BOLD fluctuations. However, the removal of the global signal is particularly controversial, and we'll return to this in a few slides. Many traditional approaches towards analyzing fMRI data, such as the general linear model that we talked a lot about in the previous course, are not relevant for resting state data due to the inherent lack of task. Instead more exploratory methods like seed analysis and independent component analysis which we'll talk about in the brain connectivity section this course are more popular. There also exist specific methods particular tailored to resting state rfMRI such as amplitude of low frequency fluctuations or ALFF, fractional ALFF, or fALFF, and regional homogeneity, ReHo. In the past few years there has been an increase in attention given to the observed anti-correlations between different resting state networks, so, negative correlations between them. Anti-correlations between the components of the default-mode and attention networks have been particularly been consistently observed. However, recently there's been a lot of debate about these findings and it's not that the global signal regression that I mentioned previously induces a bias in towards finding these anti-corredlations between resting state networks. So now often researchers perform analysis with or without the global being removed in order to see whether or not there's been an induced bias. So there is a growing subfield around the acquisition and analysis of resting state fMRI data. One of the primary benefits with this type of data is the ability to compare data across labs, so experiments don't need to be synchronized. Anyone can perform a similar type of resting state data just so you have everyone would have their eyes open and just scan them for five or ten minutes and then you can easily compare across the labs. And this has lead to the creation of large data sharing initiatives. In particular, I want to focus on the 1,000 Functional Connectomes Project, which is collecting more and more resting state data and making it openly available for researchers to use and analyze. Okay that's the end of this module and here we've been talking about resting state functional fMRIs, this is an alternative to task based fMRI and later on when we start talking about connectivity we'll return to talking more about resting state fMRI. Okay, see you then, bye.