Whole-cell segmentation of tissue images with human-level performance using large-scale data annotation and deep learning.

A principal challenge in the analysis of tissue imaging data is cell segmentation-the task of identifying the precise boundary of every cell in an image. To address this problem we constructed TissueNet, a dataset for training segmentation models that contains more than 1 million manually labeled cells, an order of magnitude more than all previously published segmentation training datasets. We used TissueNet to train Mesmer, a deep-learning-enabled segmentation algorithm. We demonstrated that Mesmer is more accurate than previous methods, generalizes to the full diversity of tissue types and imaging platforms in TissueNet, and achieves human-level performance. Mesmer enabled the automated extraction of key cellular features, such as subcellular localization of protein signal, which was challenging with previous approaches. We then adapted Mesmer to harness cell lineage information in highly multiplexed datasets and used this enhanced version to quantify cell morphology changes during human gestation. All code, data and models are released as a community resource.

The amplitude of the resting-state fMRI global signal is related to EEG vigilance measures.

In resting-state functional magnetic resonance imaging (fMRI), functional connectivity measures can be influenced by the presence of a strong global component. A widely used pre-processing method for reducing the contribution of this component is global signal regression, in which a global mean time series signal is projected out of the fMRI time series data prior to the computation of connectivity measures. However, the use of global signal regression is controversial because the method can bias the correlation values to have an approximately zero mean and may in some instances create artifactual negative correlations. In addition, while many studies treat the global signal as a non-neural confound that needs to be removed, evidence from electrophysiological and fMRI measures in primates suggests that the global signal may contain significant neural correlates. In this study, we used simultaneously acquired fMRI and electroencephalographic (EEG) measures of resting-state activity to assess the relation between the fMRI global signal and EEG measures of vigilance in humans. We found that the amplitude of the global signal (defined as the standard deviation of the global signal) exhibited a significant negative correlation with EEG vigilance across subjects studied in the eyes-closed condition. In addition, increases in EEG vigilance due to the ingestion of caffeine were significantly associated with both a decrease in global signal amplitude and an increase in the average level of anti-correlation between the default mode network and the task-positive network.

Resting-state neuronal oscillatory correlates of working memory performance.

PURPOSE: Working memory (WM) represents the brain's ability to maintain information in a readily available state for short periods of time. This study examines the resting-state cortical activity patterns that are most associated with performance on a difficult working-memory task. METHODS: Magnetoencephalographic (MEG) band-passed (delta/theta (1-7 Hz), alpha (8-13 Hz), beta (14-30 Hz)) and sensor based regional power was collected in a population of adult men (18-28 yrs, n = 24) in both an eyes-closed and eyes-open resting state. The normalized power within each resting state condition as well as the normalized change in power between eyes closed and open (zECO) were correlated with performance on a WM task. The regional and band-limited measures that were most associated with performance were then combined using singular value decomposition (SVD) to determine the degree to which zECO power was associated with performance on the three-back verbal WM task. RESULTS: Changes in power from eyes closed to open revealed a significant decrease in power in all band-widths that was most pronounced in the posterior brain regions (delta/theta band). zECO right posterior frontal and parietal cortex delta/theta power were found to be inversely correlated with three-back working memory performance. The SVD evaluation of the most correlated zECO metrics then provided a singular measure that was highly correlated with three-back performance (r = -0.73, p<0.0001). CONCLUSION: Our results indicate that there is an association between WM performance and changes in resting-state power (right posterior frontal and parietal delta/theta power). Moreover, an SVD of the most associated zECO measures produces a composite resting-state metric of regional neural oscillatory power that has an improved association with WM performance. To our knowledge, this is the first investigation that has found that changes in resting state electromagnetic neural patterns are highly associated with verbal working memory performance.

Caffeine-Induced Global Reductions in Resting-State BOLD Connectivity Reflect Widespread Decreases in MEG Connectivity.

In resting-state functional magnetic resonance imaging (fMRI), the temporal correlation between spontaneous fluctuations of the blood oxygenation level dependent (BOLD) signal from different brain regions is used to assess functional connectivity. However, because the BOLD signal is an indirect measure of neuronal activity, its complex hemodynamic nature can complicate the interpretation of differences in connectivity that are observed across conditions or subjects. For example, prior studies have shown that caffeine leads to widespread reductions in BOLD connectivity but were not able to determine if neural or vascular factors were primarily responsible for the observed decrease. In this study, we used source-localized magnetoencephalography (MEG) in conjunction with fMRI to further examine the origins of the caffeine-induced changes in BOLD connectivity. We observed widespread and significant (p < 0.01) reductions in both MEG and fMRI connectivity measures, suggesting that decreases in the connectivity of resting-state neuro-electric power fluctuations were primarily responsible for the observed BOLD connectivity changes. The MEG connectivity decreases were most pronounced in the beta band. By demonstrating the similarity in MEG and fMRI based connectivity changes, these results provide evidence for the neural basis of resting-state fMRI networks and further support the potential of MEG as a tool to characterize resting-state connectivity.

Anti-correlated networks, global signal regression, and the effects of caffeine in resting-state functional MRI.

Resting-state functional connectivity magnetic resonance imaging is proving to be an essential tool for the characterization of functional networks in the brain. Two of the major networks that have been identified are the default mode network (DMN) and the task positive network (TPN). Although prior work indicates that these two networks are anti-correlated, the findings are controversial because the anti-correlations are often found only after the application of a pre-processing step, known as global signal regression, that can produce artifactual anti-correlations. In this paper, we show that, for subjects studied in an eyes-closed rest state, caffeine can significantly enhance the detection of anti-correlations between the DMN and TPN without the need for global signal regression. In line with these findings, we find that caffeine also leads to widespread decreases in connectivity and global signal amplitude. Using a recently introduced geometric model of global signal effects, we demonstrate that these decreases are consistent with the removal of an additive global signal confound. In contrast to the effects observed in the eyes-closed rest state, caffeine did not lead to significant changes in global functional connectivity in the eyes-open rest state.

Accurate reconstruction of temporal correlation for neuronal sources using the enhanced dual-core MEG beamformer.

Beamformer spatial filters are commonly used to explore the active neuronal sources underlying magnetoencephalography (MEG) recordings at low signal-to-noise ratio (SNR). Conventional beamformer techniques are successful in localizing uncorrelated neuronal sources under poor SNR conditions. However, the spatial and temporal features from conventional beamformer reconstructions suffer when sources are correlated, which is a common and important property of real neuronal networks. Dual-beamformer techniques, originally developed by Brookes et al. to deal with this limitation, successfully localize highly-correlated sources and determine their orientations and weightings, but their performance degrades at low correlations. They also lack the capability to produce individual time courses and therefore cannot quantify source correlation. In this paper, we present an enhanced formulation of our earlier dual-core beamformer (DCBF) approach that reconstructs individual source time courses and their correlations. Through computer simulations, we show that the enhanced DCBF (eDCBF) consistently and accurately models dual-source activity regardless of the correlation strength. Simulations also show that a multi-core extension of eDCBF effectively handles the presence of additional correlated sources. In a human auditory task, we further demonstrate that eDCBF accurately reconstructs left and right auditory temporal responses and their correlations. Spatial resolution and source localization strategies corresponding to different measures within the eDCBF framework are also discussed. In summary, eDCBF accurately reconstructs source spatio-temporal behavior, providing a means for characterizing complex neuronal networks and their communication.