There is a huge body of literature on resting state fMRI analysis.

However, a good way to analyze BOLD-correlates of different "global brain states" is to add a simultaneous EEG mesurement to your design. Then you have the possibility to  assess different  e.g. vigilance stages by EEG (http://research.uni-leipzig.de/vigall/) and take the resulting  EEG-based time series as predictors for your fMRI analysis.

https://www.researchgate.net/publication/23706744_EEG-vigilance_and_BOLD_effect_during_simultaneous_EEGfMRI_measurement

Article EEG-vigilance and BOLD effect during simultaneous EEG/fMRI m...

There are an abundance of tools and methods for this. We have had good results with looking at mean connectivity measures at both the single subject level and the group level see

http://online.liebertpub.com/doi/abs/10.1089/brain.2014.0244

http://www.jneurosci.org/content/31/36/12972.full

Here are some methods you could use which was published early this year (citation below):

Use a high-dimensional model-order ICA on resting-state scans using Group ICA fMRI Toolbox (GIFTv3.0a; http://mialab.mrn.org/software/gift/). Prior to the first and second PCA data reductions (1st at c=100; final at c=75), which were computed on each subjects׳ dataset before doing the group PCA, the time-series at each voxel were scaled to percent signal change using the intensity normalization function in GIFT. This option was selected based on a recent study which showed that this pre-processing method maximizes the detectability of individual differences in functional connectivity measures at the group-level ( Khullar et al., 2011; Erhardt et al., 2011). The infomax ICA algorithm, which incorporates nonlinearities in the transfer function to capture higher-order moments in the BOLD resting-state fluctuations ( Bell and Sejnowski, 1995), was then used to detect the independent source components in the data. ICA was repeated 15 times using ICASSO toolbox in order to estimate the algorithmic and statistical reliability of the ICA decomposition ( Himberg, Hyvärinen, and Esposito, 2004). Back-reconstruction was then run using group ICA (GICA3), with component spatial maps of functional connectivity scaled to represent the original data units in percent signal change.

Hunter, M. A., Coffman, B. A., Gasparovic, C., Calhoun, V. D., Trumbo, M. C., & Clark, V. P. (2015). Baseline effects of transcranial direct current stimulation on glutamatergic neurotransmission and large-scale network connectivity. Brain Research, 1594, 92–107. http://doi.org/10.1016/j.brainres.2014.09.066

Lei Gao I believe you want to obtain something like percent signal change in resting state fMRI - not connectivity, or proxies, etc. but actual BOLD.  The problem is you do not have an implicit contrast in the resting state as you would in task related data.  Perhaps look at ALFF or fALFF type of analyses?