Recent studies employing rapid sampling techniques have demonstrated that the resting state fMRI (rs-fMRI) signal exhibits synchronized activities at frequencies much higher than the conventional frequency range (<0.1 Hz). However, little work has investigated the changes in the high-frequency fluctuations between different resting states. Here, we acquired rs-fMRI data at a high sampling rate (TR = 400 ms) from subjects with both eyes open (EO) and eyes closed (EC), and compared the amplitude of fluctuation (AF) between EO and EC for both the low- and high-frequency components. In addition to robust AF differences in the conventional low frequency band (<0.1 Hz) in visual cortex, primary auditory cortex and primary sensorimotor cortex (PSMC), we also detected high-frequency (primarily in 0.1–0.35 Hz) differences. The high-frequency results without covariates regression exhibited noisy patterns. For the data with nuisance covariates regression, we found a significant and reproducible reduction in high-frequency AF between EO and EC in the bilateral PSMC and the supplementary motor area (SMA), and an increase in high-frequency AF in the left middle occipital gyrus (MOG). Furthermore, we investigated the effect of sampling rate by down-sampling the data to effective TR = 2 s. Briefly, by using the rapid sampling rate, we were able to detect more regions with significant differences while identifying fewer artifactual differences in the high-frequency bands as compared to the down-sampled dataset. We concluded that (1) high-frequency fluctuations of rs-fMRI signals can be modulated by different resting states and thus may be of physiological importance; and (2) the regression of covariates and the use of fast sampling rates are superior for revealing high-frequency differences in rs-fMRI signals.

Resting-state fMRI studies have increasingly focused on multi-contrast techniques, such as BOLD and ASL imaging. However, these techniques may reveal different aspects of brain activity (e.g., static vs. dynamic), and little is known about the similarity or disparity of these techniques in detecting resting-state brain activity. It is therefore important to assess the static and dynamic characteristics of these fMRI techniques to guide future applications. Here we acquired fMRI data while subjects were in eyes-closed (EC) and eyes-open (EO) states, using both ASL and BOLD techniques, at two research centers (NIDA and HNU). Static brain activity was calculated as voxel-wise mean cerebral blood flow (CBF) using ASL, i.e., CBF-mean, while dynamic activity was measured by the amplitude of low frequency fluctuations (ALFF) of BOLD, i.e., BOLD-ALFF, at both NIDA and HNU, and CBF, i.e., CBF-ALFF, at NIDA. We showed that mean CBF was lower under EC than EO in the primary visual cortex, while BOLD-ALFF was higher under EC in the primary somatosensory cortices extending to the primary auditory cortices and lower in the lateral occipital area. Interestingly, mean CBF and BOLD-ALFF results overlapped at the visual cortex to a very small degree. Importantly, these findings were largely replicated by the HNU dataset. State differences found by CBF-ALFF were located in the primary auditory cortices, which were generally a subset of BOLD-ALFF and showed no spatial overlap with CBF-mean. In conclusion, static brain activity measured by mean CBF and dynamic brain activity measured by BOLD- and CBF-ALFF may reflect different aspects of resting-state brain activity and a combination of ASL and BOLD may provide complementary information on the biophysical and physiological processes of the brain.

The 2016 WHO classification of central nervous system tumors has included four molecular subgroups under medulloblastoma (MB) as sonic hedgehog (SHH), wingless (WNT), Grade 3, and Group 4. We aimed to develop machine learning models for predicting MB molecular subgroups based on multi-parameter magnetic resonance imaging (MRI) radiomics, tumor locations, and clinical factors. A total of 122 MB patients were enrolled retrospectively. After selecting robust, non-redundant, and relevant features from 5,529 extracted radiomics features, a random forest model was constructed based on a training cohort ( n = 92) and evaluated on a testing cohort ( n = 30). By combining radiographic features and clinical parameters, two combined prediction models were also built. The subgroup can be classified using an 11-feature radiomics model with a high area under the curve (AUC) of 0.8264 for WNT and modest AUCs of 0.6683, 0.6004, and 0.6979 for SHH, Group 3, and Group 4 in the testing cohort, respectively. Incorporating location and hydrocephalus into the radiomics model resulted in improved AUCs of 0.8403 and 0.8317 for WNT and SHH, respectively. After adding gender and age, the AUCs for WNT and SHH were further improved to 0.9097 and 0.8654, while the accuracies were 70 and 86.67% for Group 3 and Group 4, respectively. Prediction performance was excellent for WNT and SHH, while that for Group 3 and Group 4 needs further improvements. Machine learning algorithms offer potentials to non-invasively predict the molecular subgroups of MB.

No abstract