Figure 4: Results from analysis of a testing dataset. Our self-supervised deep learning shows improved tSNR (left) compared to conventional reconstruction. The angular maps (middle), and the correlation between time-series data and the pRF model fit thresholded at 0.2 (right), both masked to occipital love, show no meaningful differences, suggesting no loss in functional spatial precision or subtle temporal effects with the deep learning method.

Figure 1: A schematic of the self-supervised approach used in this study, which does not require fully-sampled data. The unrolled neural network incorporates the encoding operator via data consistency steps. A ResNet is used for non-linear regularization. The data split ($$$\Theta,\Lambda$$$) of the acquired k-space locations ($$$\Omega$$$) allow training without fully-sampled data, which has been a main hindrance for DL reconstruction for fMRI.

Animated Figure 3. Standard Looping Star fMRI (left) vs high-temporal resolution (middle) and high-spatial resolution (right) extreme Looping Star fMRI. Performing fMRI with a motor task can be done with either higher temporal resolution (middle) or higher spatial resolution (right) that preserves structural detail compared to standard Looping Star (left). An animated GIF (5x speedup) of the temporal volumes is shown (top row) and the corresponding thresholded motor task activation maps (bottom row) analyzed using FSL FEAT. Only 1 second is shown due to file size limits.

Figure 5. Demonstration of Extreme Looping Star for a visual task at a different site. The activation map clearly shows activation in the visual cortex at a TR of 0.155s and 1280 volumes with an effective sub-Nyquist sampling factor per volume of 0.025. This is 10x finer temporal resolution than original Looping Star.

Fig 2: Whole-brain single-subject GLM results for SE BOLD and dfMRI. In order to find activation without making assumptions about the shape of the response functions, a Finite Impulse Response (FIR) set was used. Significance was assessed with F-tests on the FIR regressors. The F-statistic maps were converted to the displayed z-statistic maps, which were additionally truncated at 2.3. Activation in visual and motor cortices was observed for all three modalities.

Fig 4: Group averages of Fisher-transformed FC matrices from SE-BOLD (A) and b=0.2/1 ADC (B) at 3T and 7T. (C) Correlation of FC strength derived from SE-BOLD vs dfMRI. Positive correlations were somewhat less pronounced in dfMRI compared to BOLD (r ~ 0.7 for both 3T and 7T), but anti-correlations were attenuated preferentially in dfMRI, particularly at 3T (r=0.16).

Figure 5. Activation maps resulting from the flashing checkerboard experiment by the shot-combined SENSE reconstruction (top) and the proposed reconstruction (bottom). In the sagittal view, activation in the calcarine sulcus is better characterised in the proposed reconstruction, and overall improved sensitivity to activation and higher peak z-stats are observed in the proposed reconstruction.

Figure 4. The reconstruction results of the 1.8mm in-vivo data. A. Mean tSNR across the whole brain for different trajectory choices by the conventional shot-combined SENSE reconstruction and the proposed reconstruction for the original R_net=2x1 data without retrospective under-sampling. B. tSNR(mean value shown on the bottom left), temporal mean and standard deviation maps by the conventional shot-combined SENSE reconstruction(left) and the proposed reconstruction(right) for the seg-CAIPI(8,2) trajectory at R_net=2x2, R_net=2x4 and R_net=2x6.

Figure 4. B₀ field maps and gradient echo EPI showing the impact of passive and active shimming, plus the combination of the two. Anatomic images show areas of signal dropout in EPI. The yellow arrows show voxels where the ear shim reduces signal dropout. The red arrows show a region of severe B₀ artifacts that is improved by the active + passive shimming, but not fully mitigated. The green arrows show a region where the ear shim “overshoots” and actually worsens the shim in a region anterior to the B₀ hotspot. Both global and dynamic MC shimming are effective in mitigating this overshoot field.

Figure 1. Panel A: Diamagnetic pyrolytic graphite was blended into a silicon putty to create a passive earplug shim. Spherical pieces were placed in the lobe adjacent to the elongated shim, which was inserted into the ear canal.

Panel B: Thermal images were obtained before and after scanning ear shim material in a receive array head coil. We conducted imaging of the shims at baseline and after running a turbo spin echo at 99% of the SAR limit for 10 minutes and gradient slew test with maximum slew rate EPI for 10 minutes. No significant change in temperature was observed.

Functional images as t-maps (target > surround) thresholded at t ≥ 5.7 for a single NORDIC processed run (left most panel) and for 1, 3 and 5 Standard processed runs combined, for subject 1 (S1, top row) and subject 2 (S2, lower row). The data were with a 2D GE 0.8 mm iso voxels (iPAT 3; MB2; TR 1350) using a 12 seconds on and 12 seconds off block design where 2 flickering checkerboard (target and surround) were presented 3 times each per run.

Figure 4: NORDIC denoising on 3T fMRI using the 3T HCP protocol. T-maps with t value ≥ for the contrast target (red) > surround (blue) for a representative slice in the visual cortex (top), and on inflated cortex (bottom). Approximately four runs with standard reconstruction (~10 min of data) are required to achieve the extent of activation comparable a single NORDIC run (~ 2.5 min of data). The right-most panel shows the t-value distribution for NORDIC, Standard, and Standard reconstructions with 2 mm FWHM spatial smoothing, within a the retinotopic representation of the target in V1.

Figure 3: Spectrograms of uncorrected versus corrected magnitude data. The location of the cardiac and respiratory frequency bands are indicated in the uncorrected magnitude image (left) with red and black horizontal arrows, respectively. While some residual cardiac and respiration power remained after the RETROICOR(PMU) and PESTICA correction (red and black vertical arrows in the central two figures, respectively), PREPAIR showed effective removal of these frequencies.

Figure 4: Average variance improvement over subjects (in percentage) for protocols with different TRs after correction with both respiratory and cardiac regression. PREPAIR (red) achieved similar image variance reduction to RETROICOR(PMU) (green) and outperformed in PESTICA (blue) for all protocols.

Figure 2: Spike model simulations. (Left) Heatmap of the regularization paths of the activity-inducing signal $$$\mathbf{s}$$$ estimated with PFM and TA as a function of $$$\lambda$$$ (increasing number of iterations in x-axis), whereas each row in the y-axis shows one time point. Vertical lines denote iterations corresponding to the Akaike and Bayesian Information Criteria (AIC and BIC). (Right) Estimated activity-inducing (blue) and activity-related (green) signals when $$$\lambda$$$ is set based on BIC. All estimates of $$$\mathbf{s}$$$ are identical, regardless of SNR.

Figure 4: (Row 1) Regularization paths of the estimated activity-inducing signal $$$\mathbf{s}$$$ (spike model - left) and innovation signal $$$\mathbf{u}$$$ (block model - right); (Row 2) activity-inducing, innovation and activity-related (fit, $$$\mathbf{x}$$$) signals when $$$\lambda$$$ is chosen based on BIC, or (Row 3) based on convergence of residuals to have same variance as MAD estimate of noise; (Row 4) Corresponding curves of BIC and AIC, where the vertical lines indicate the three options to select $$$\lambda$$$ (BIC, AIC and Converged/MAD).

Figure 1. F-map of the task effect. AC values are significantly different during different cognitive tasks in the majority of the gray-matter cortex.

Figure 2. Paired comparison of the AC patterns between the tasks. Cognitively demanding tasks like working memory and mathematical computations significantly reduces the AC values in the brain regions associated with those tasks.

Layering metrics generated in LayNii:

The top row shows an application with a synthetic 2D image. The middle row shows the empirical layers from Ding et al. (2016) (0.2 mm iso.). The bottom row shows BigBrain (0.1 mm iso., native space) (Amunts et al. 2013) with cortical borders provided in Wagstyl et al. (2020). The equi-distant metric is shown in the middle column and equi-volume metric is shown in the right column for each image type. The arrows highlight areas where the equi-distant and the equi-volume metric differ considerably.

Estimating columnar units in voxel space with LayNii:

Panel A) describes the algorithm of estimating columnar distances in the example of a cat visual cortex.

Panel B) depicts cortical unfolding in LayNii. The orthogonal coordinates of columnar distances and cortical depths are used to map the signal intensity on a new spatial grid and then LayNii writes it out as NIFTI files.

Panel C) depicts an application for topographic mapping in the human somatosensory system of depicting digit representation in the anterior bank of the central sulcus acros laminar and columnar directions.

Recorded hand motion during an fMRI scan. The bottom figure shows a zoomed view of the first section. Different motion states are highlighted in the figures. The trigger signal from the MR system is also shown in the figure (orange line).

Activation map of subject 1 obtained by GLM using motion states (a) (i) and (ii), (b) (i), (ii) and (iii), (c) (i), (ii), (iii) and (iv). 2d GRE-EPI sequence was used with following parameters; TR = 0.5sec, TE = 24.6ms, flip-angle = 40°, 1.1mm isotropic resolution, 18 slices, 143mm x 143mm FOV, GRAPPA = 3, MB = 2, 490 TRs. (d) Signal time series and model predictions of subject 1. (e) Recorded hand motion, signal timeseries and model predictions of subject 2. The mean signal time series calculated within motor cortex with Z>5. The figure legend on the bottom right applies for both (d) and (e).

Figure 1. R₂’ vs. vessel radius: comparison of cylindrical and spherical perturbers for gradient echo (GE), spin echo (SE) and asymmetric spin echo (ASE). Uses CBV=2%, Y=60%, Hct=35.7%.

Figure 2. R₂’ vs. CBV: comparison of cylindrical and spherical perturbers for gradient echo (GE), spin echo (SE) and asymmetric spin echo (ASE). Uses radius R=15μm, Y=60%, Hct=35.7%.

(a) Highest beta estimates in each ROI for the three tasks, pooled over coils. Larger beta estimates without prescan normalize in motor and visual cortex, but no difference in auditory cortex and the opposite pattern in thalamus. (b) Beta estimates for the two coils pooled over tasks. The same pattern holds for both coils. (mot = motor, aud = auditory, vis = visual task, ON/OFF = with/without prescan normalize)

Normalized tSNR maps for the 20-channel head coil (a,b) and the 64-channel head coil (c,d) with (ON) and without (OFF) prescan normalize.

Fig 3. a: Signal change as a function of vessel size calculated in time steps of 0.4 ms along a multi-echo spin echo train with perfect 180° refocusing pulse at 7T with echo spacings of 20 ms. The (gradient echo-like) profiles immediately before and after later refocusing show increasingly less sensitivity to larger vessels. b: Mean signal changes along GRASE echo trains at 7T using echo spacings of 15, 20, 35 with variable refocusing flip angles that target an echo amplitude of 0.1 and 0.07 M0 (6, 8) and 48 ms echo spacing with constant refocusing flip angles of 165° (9) as well as 90° and 180°.

Fig. 2. Top row: Size of vessel radius with maximal signal change DM0 as a function of echo time and field strength and refocusing flip angles. For small refocusing pulses the vessel radius with maximal signal change increases with echo number. Bottom row: mean signal change (averaged across vessel radius) as a function of constant refocusing flip angles from 20° to 180°, and for varying refocusing flip angles with target echo amplitudes of 0.1, 0.2 and 0.3 M0. These flip angles were calculated for field strength dependent relaxation times according to the algorithm given by Busse (10).

Figure 3: Output of fMRI analysis at group level and first level across participants for the FID and echo images. Significant (cluster level p_FWE < 0.05) activity was identified in the visual cortex in both cases, with statistics listed below the figures. The tSNR maps of Looping Star are also illustrated on the right.

Figure 2: Output images from Looping Star acquisition techniques in a single participant. From left to right, the high resolution anatomical image, wrapped phase images, unwrapped frequency map, residual difference field, QSM maps and fMRI raw images are shown. Red arrows indicate clarity of ventricles in fMRI acquisition. QSM image range -0.1ppm to 0.1ppm.

Figure 4: (A): The three segments at different locations as shown in yellow boxes together with the FOV used in the acquisition. The lower and upper blocks covered the visual cortex and motor cortex, respectively. (B): A composite SMSeg image of (A) used in fMRI study (10 averages of images obtained during the baseline). (C): The composite activation map from all three segments showed simultaneous activations in the visual (the lower segment) and motor cortices (primary motor cortex and supplementary motor area in the upper segment).

Figure 2: In (A), three obliquely distributed segments were excited simultaneously and projected into one composite “slice”. Each column in the right panel shows two representative axial composite slices of the phantom from the SMSeg EPI sequence without in-plane acceleration (B, E) and with 4-fold acceleration (C, F), together with the conventional 2D EPI with 4-fold acceleration at the location corresponding to the central segment (D, G). The SNRs of images (C) and (D) at the central segment were 50.9 and 46.3, respectively.

Figure 1. Schematics showing that narrow-band velocity selectivity allows ultra-fast perfusion imaging with high SNR efficiency, while conventional VSASL (wide-band labeling) may suffer from reduced SNR efficiency due to multiple labeling on the same bolus.

Figure 3. The velocity-selective profiles under the label condition for both modified rect-VSI and sinc-VSI labeling. Under the control condition, the magnetization response was the same across all velocities (not shown), of the same value at V = 0 under the label condition.

Single-subject ALFF & ReHo maps, acquired from the same subject depicted in Fig. 1. White arrows highlight a region of strong recovery and a node that was recovered with the SPSP pulse.

Regions of significant difference (p < 0.05) between the compensated & uncompensated ALFF & ReHo maps, calculated across all subjects.

Figure 4. Sequence comparison with FSNR: explainable variance (EV). A) Median EV in a cortical (left) and ATL mask (right). The iPAT1 single-echo sequence (blue) shows higher EV in the whole brain than the other sequences, but the multi-echo sequence (green) shows higher EV in ATL. POCS reconstruction also improves EV in the ATL. B) EV on the cortical surface with LB ON. The multi-echo sequence (green) shows higher EV than the other sequences in temporal cortex.

Figure 1. Sequences tested. Each row indicates the metric used to assess the sequence (temporal SNR, functional SNR), in-plane acceleration factor (iPAT), multiband acceleration factor (MB), partial fourier factor (PF), TR, and echo times (TE1-3). Sequences were collected with LeakBlock on (LB ON), and were retro-reconstructed with either LeakBlock off (LB OFF), LeakBlock on and POCS (LB ON + POCS), or POCS only (LB OFF + POCS). PERFGR: gradients in performance mode.

Maps of the mean sensitivity, across subjects, for images acquired along axial and sagittal directions planes. CNR is comparable for the two protocols

Reproducibility in gray matter (on the left panel) and in Cord (in the right panel). Reproducibility is greater for axial acquisition protocol than sagittal one scanning.

Fig. 1: Schematic diagram of the processing pipeline. The fMRI data of each functional run is first submitted to motion and slice timing correction. Then, geometric distortions are corrected using AP-PA or FM approaches. Subsequently, additional pre-processing steps are performed on the corrected and uncorrected data, followed by the three different data analyses: 1) registration into structural data, 2) identification of resting-state networks, and 3) mapping of task-related brain regions of interest.

Fig. 3: Ten group RSNs identified on (left) uncorrected, (middle) AP-PA corrected, and (right) FM-corrected fMRI data for the first run of the biological motion perception task. The RSN templates (in red-yellow) from ⁸ are superimposed with the spatial independent components (in blue-light blue) selected for each RSN template, according to their Dice coefficient (also shown above each RSN).

Figure 5. Example parameter maps for all subjects scanned. OEF estimates are approximately uniform in grey matter. High OEF estimates in the white matter are likely due to a lack of perfusion signal.

Figure 3: A) Preprocessed fMRI and estimated neuronal-related Time-series in a voxel in the tongue motor area (see cross in map); B) single-trial GLM maps ($$$p<0.001$$$)and LR+MV-SPFM activation maps and C) maps of the three estimated low-rank components. The colour bands in the plots with time-series denote the timing of the different conditions.

Figure 1: Simulation results. A) Example of simulated signals for different SNR conditions; B) ROC curves for the estimation of the neuronal-related signal with: SPFM using BIC (SPFM BIC), SPFM with no LR estimation and no spatial regularization (SPFM, $$$ \rho=1$$$), MV-SPFM with no LR estimation (MV-SPFM, $$$ \rho=0.8$$$), the LR+MV-SPFM algorithm with only the L1-norm (LR+SPFM, $$$ \rho=1$$$), and the LR+MV-SPFM algorithm ($$$ \rho=0.8$$$). C) Estimation error of the LR components for different ratios of BOLD/total number of voxels.

Group level activation maps for the four cases; yellow arrows: primary sensorimotor cortex (M1, S1, M_full-BW_high: z=13.09±3.7, M_full-BW_low: z=11.76±3.3, M_crop-BW_low: z=13.03±3.4, M_crop-BW_high: z=14.23±3.9) blue arrows: superior parietal lobule (SPL-L), green arrows: primary somatosensory cortex (S1-L).

The timing diagram of the ADC (red) and frequency encoding gradient (blue) for each of the four acquisition schemes, Full matrix/high BW (M_full-BW_high), full matrix/low BW (M_full-BW_low), cropped matrix/low BW (M_crop-BW_low), cropped matrix/high BW (M_crop-BW_high).

Figure 1. Spectral rigidity for controls (nopain) vs. those with osteopathic pain taking duloxetine. Solid lines indicate group mean rigidity, and shaded regions correspond to 99% percentile bootstrapped intervals for each group. Vertical axes vary to better depict overlap.

Figure 2. Level variance for controls (nopain) vs. those with osteopathic pain taking duloxetine. Solid lines indicate group mean rigidity, and shaded regions correspond to 99% percentile bootstrapped intervals for each group. Vertical axes vary to better depict overlap.

Fig. 5. Human Subject: A-B, Activation maps computed from Pearson correlation coefficient with a threshold of 0.45. C-D, Compare time courses for two activation pixels in the visual cortex

Fig. 2. 1DCNN architecture, The input of the network is a 15s long 1D OSSI signal and its output is the corresponding 1D BOLD signal. 95% of the simulated BOLD and OSSI time courses were used for training while the remaining 5% were set aside for testing. L₁ loss, ADAM optimizer with a learning rate of 0.007 was used for training the network.

Figure 5: Application of the present method to an in vivo high-speed resting-state fMRI scan and comparison with full-width, whole band correction of motion-related signal changes with seeds in the auditory network (AUN), the default mode network (DMN), and the visual network (VSN). Data were acquired using multi-slab Echo Volumar Imaging with TR: 246 ms, isotropic voxel size: 4 mm, number of time points: 3000, multi-band factor: 2, in-plane GRAPPA acceleration: 3. Correlation threshold: 0.3.

Figure 3: Dependence of spectrally segmented regression of the noised model (simulated resting-state fMRI data with noise from in-vivo scan motion parameters added) on the number of spectral bands of the regression vectors. False positive correlations decrease with increasing spectral segmentation. Color bar shows correlation range between -1 and 1 and correlation threshold: 0.1.

Figure 5: A) Sagittal slice showing activations from a representative subject at the left pars orbitalis. The activation maps correspond to: (left) RAW analysis, (middle) OLS phase-based denoising, (right) ODR phase-based denoising. B) Group-level effects of denoising in the activation maps at the left pars orbitalis. The maps depict significant differences using a mixed effects model between the analysis approaches: (top) RAW vs. ODR, (middle) RAW vs. OLS, (bottom) OLS vs. ODR.

Figure 3: R² maps of both OLS and ODR fitting methods, and magnitude (HPF 0.25 Hz) temporal standard deviation (tSTD) map shown in the sagittal plane for a representative subject. The overlaid mask in green indicates the segmented veins from the T1w/T2w ratio images.

Fig. 1. Three stages of adapted NPE for fMRI. (1): The N subjects are applied with SVD and the correlation are used to form the adjacency graph. The green strip denotes a spatial eigenvector from sub#1. Each dot denotes a connection, and the darker colors denotes stronger connection. (2): The red strips and dots are qualified components forming a neighborhood with green one, while white dots are disqualified components with weak connection. (3): The Linear approximation is employed to compute the weight with well-constructed neighborhood and followed with projection to obtain results.

Fig. 4. The consistency analyses for spatial-temporal ICs of Group ICA under model order 30. (A) The lines are the sorted correlation of paired ICs in lower MO, 10(blue), 15 (green), 20 (red), and 25 (light blue) with IC in MO 30. The horizonal black dotted line is set as 0.8 to quantify reproducible ICs from NPE (solid line) and PCA (dotted line). (B) The bar height denotes IC numbers in different consistency index ranges (<0.7, 0.7-0.8, 0.8-0.9, >0.9) for NPE (red) than PCA (light blue). (C) Correlation maps of paired ICs from MO 10 to 30 with ICs in MO 30.

Fig. 4, The T map and BOLD Index mapping under the threshold of p=0.05 and p=0.01. In order to simplify the display, for the BOLD Index, only the positive activation regions are displayed. The parameter K in BOLD Index equals 1. If the BOLD Index is bigger than 0.5 which means the voxel is a BOLD signal, and if the BOLD Index is smaller than 0.5 which means the voxel is a non-BOLD signal/false positive. So the blue color in c),d) labels non-BOLD/false positive regions. There are more blue regions in c) under p<0.05 than in d) under p<0.01.

Fig. 5, The T map from standard analysis and BOLD index in smaller voxel size 2*2*3mm³. The upper part shows the results of the motor cortex and the lower part shows the results of the visual cortex. a) shows the T map under the threshold of p=0.05; b) provides the anatomical structure to show where the activation is; c) displays the BOLD index inside the mask from the T map. d), e) and f) are the magnified interest region from a), b) and c) respectively. The parameter K=4 is selected for BOLD Index to display the detailed activation patterns.

Figure 2: Thresholded statistical maps for each of the early, late and canonical responses, for both the magnitude (upper row) and phase time series (lower row). The shaded square illustrates the hand knob ROI used as an exclusive mask for the TDM and GLM analysis as well as image realignment.

Figure 1: The two waveforms (early and late) extracted from the magnitude images, using the TDM method. The profiles are seen to differ slightly on their rising edge, and at the post stimulus undershoot.

Figure 2 – Activation maps for 1mm Spin Echo (SE) and gSlider Super Resolution (SR) data with different levels of regularization (Lambda).

5×-gSlider with ‘slice-phase dither’ encoding to provide highly independent basis, while maintaining high image-SNR in each individual slab acquisition. The DIST RF pulses are shown in the left column, with real and imaginary parts shown in black and green respectively. The corresponding slab profiles are shown in the middle column, each with a π phase dithering applied to a different sub-slice.

Figure 2: Single-subject perfusion tSNR maps and functional activation maps overlaid on the mean 3D GRASE images for PASL CAIPI (A,C) and PCASL CAIPI (B,D). Note the identical colorbar scaling for A and B and C and D, respectively.

Figure 3: Group mean and standard deviations of the grey matter control tSNR (A) and perfusion tSNR (B) efficiencies across labeling approaches and acceleration schemes.

Figure 3. Comparison of group functional contrast maps between the distortion-corrected forward (DF), the reverse (DR), and the combined EPI images (DW) on a 3D inflated surface model. Group average maps of GLM T-statistics (A) and signal percentage changes (B) are present for the comparison.

Figure 2. Comparison of cortical coverage ratio between the distortion-corrected forward (DF), the reverse (DR), and the combined EPI images (DW). (A) The cortical coverage for six subjects is visualized on the inferior view of 3D inflated surface model. (B) The coverage ratios at the local areas of both hemispheres are present as the mean and standard deviation. The specific locations presenting apparent coverage ratio differences are indicated by arrows and dashed-rectangular boxes in (A).

Functional results including activation maps, temporal SNR maps, and time courses. The proposed approach has well preserved functional signals with similar activation map and time course to the fully sampled case.

Because the data shared images were reconstructed by combining k-space of every 10 slow time points, the time-series of data shared images were generated by repeating each data shared image for 10 times along slow time.

The proposed network that reconstructs nc = 10 fast time images as a sequence. The network combines U-Nets with a recurrent layer. The recurrent layer “fw” is located at the bottleneck of the U-Net, and takes learned representations from the U-Net encoder and hidden states hi (i = 1, 2, ... , nc) from the previous fast time to generate the hidden state for the next fast time frame. xi denotes zero-filled fast time image, di denotes data shared image, and yi denotes two-channel (real and imaginary) output image from the network.

Figure 1) Cerebellar cortex microcircuit model. Populations (p) = GrC, GoC, MLI, PC. Connections (c) = mf-GrC, mf-GoC, GrC-GoC, GoC-GrC, GoC-GoC, GrC-MLI, MLI-MLI, GrC-PC, MLI-PC. The external input 𝜈_input[Hz] is relayed by mossy fibers (mf) to GrC and GoC. The output activity 𝜈_PC projects to the Deep Cerebellar Nuclei (DCN). Per p or c: mp=conductance; 𝜈_p=firing rate[Hz]; K_c=connections probability*cells numbers; Q_c = quantal synaptic conductance, 𝜏_c=synaptic time decay.

Figure 2) Pipeline to compute the transfer function (F). A): Color-map showing from max(yellow) to min(blue) parameters values used for F fitting. A1) Population conductances assume high values for high excitation combined with low inhibition. A2) PC input conductance depending on connected populations conductances. A3) PC membrane voltage properties: μ_V follows the same conductances trend. σ_V is higher for low excitation and high inhibition. 𝜏_V is higher for low excitation-high inhibition. B): Cerebellum Mean-Field Model describing activity evolution of each populations

Figure 2. The architecture of the multi-task deep neural network.

Figure 3. The normalized z-scores for facial expression and identity at each DNN layer.

Figure 1. Distributions of T-values. T-values were extracted from the Target ROI (Center > Surround or Faces > Scrambled) derived from Offline data. The t-values obtained with NORDIC (Orange, dashed) processed data is comparable to the effects of an additional 1 or 1.5 voxels FWHM gaussian smoothing. While temporal smoothing did increase T-statistics for Dataset 1, for the fast event-related design (Dataset 3) this led positive and negative t-values, an effect not found in NORDIC data.

Figure 4. NORDIC leads to higher cross-validation performance. A) Exhaustive cross validation performance when training on using only one run. R² is shown over voxel inclusion threshold from full model. Examples of single voxel, single run FIR model estimates are shown. B) K-Fold training was repeated with more runs. R² performance (higher is better) within mask of voxels that explained 15% of variance, for increasing number of training datasets. Error bars are standard error over permutations.

Figure 1: Parcel number vs. entropy z-score for motor task. Task-relevant brain regions are indicated. Note lines at z = +/- 1.96, indicating thresholds for parcels w/ statistical significance.

Figure 3: Parcel power vs. entropy for motor (left; r = -0.685) and emotion (right; r = -0.670) scans. Negative correlation held for all task-state scans, with correlation coefficients ranging from r = -0.670 (emotion task) to r = -0.820 (relational processing task).

Figure 2. Single subject (1) maps for six and four task blocks, with $$$p$$$ held fixed at $$$1\times{10}^{-6}$$$ (full scans: $$$180$$$ degrees of freedom $$$[DOF]$$$, $$$t=5.07$$$; truncated scans: $$$120$$$ $$$DOF$$$,$$$t=5.16$$$). Maps are identically slice-locked to localize the visual word form area (VWFA; blue arrows, top left). At top left, full scan, control; top right, full scan, denoised; bottom left, truncated scan, control; bottom right, truncated scan, denoised. Volumes of active voxels within an ROI encompassing the VWFA are overlain (vol_ROI).

Figure 1. At left, plot in red of VWFA ROI-specific ratios of LLR to control group mean tSNR with error bars showing standard deviation, with overlain plot of whole-brain mean tSNR ratios in cyan. At right, an individual subject (from Figures 2-3) example of an ROI enveloping the visual word form area used for mean tSNR and for cluster volume tracking in Figure 2. LLR significantly increases mean tSNR both in the VWFA ROI and globally.

Figure 2. Neuroradiologist consensus threshold of motor cortex for individual subject. At top left, control (conventional) activation map at consensus threshold; bottom right, LLR-denoised map at consensus threshold; bottom left, control at consensus threshold of denoised; top right, denoised at consensus threshold of control. LLR data are thresholded substantially higher with preservation of bilateral motor cortical activation, lost by control data at denoised threshold (white arrow).

Figure 1. Neuroradiologist consensus threshold of cerebellum for individual subject. At top left, control (conventional) activation map at consensus threshold; bottom right, LLR-denoised map at consensus threshold; bottom left, control at consensus threshold of denoised; top right, denoised at consensus threshold of control. LLR data are thresholded substantially higher with preservation of bilateral cerebellar activation, lost by control data at denoised threshold (white arrow).

Figure 3. The mean fMRI timeseries from 5 seconds prior and 25 post stimuli period plotted separately for each stimuli duration (0.5 second stimuli in blue, 1 second stimuli in orange, 2 seconds stimuli in green, 3 seconds stimuli in red and 4 seconds stimuli in purple). As expected, the visual cortex shows a perfectly linear response in respect to the length of the stimulus. Similar linearity patterns are shown within DMN regions except with 0.5 and 1 second stimuli.

Figure 2, Spatial pattern of the responsive voxels for positive and negative BOLD response in the DMN and primary visual cortex overlaid on three orthogonal and most informative slices.

Figure 1. Quitting smokers' brain activation increased areas (fALFF) than that before smoking cessation treatment. A: Brain activation in smokers increased after cessation treatment when saw smoking images. B: Brain activation in smokers increased after cessation treatment when saw nonsmoking pictures.

Figure 2. Brain activation areas in smokers when they saw smoking pictures compared with non-smoking pictures before smoking cessation treatment.

Functional connectivity (ROI-based analysis, seed ROI – MPFC, 3D view)

Functional connectivity (ROI-based analysis, seed ROI – MPFC, plain view, axial).

Figure 3: The Experimental-Naturalistic Spectrum. Movie-watching H values reflect a more fractal state than resting-state. Previous findings using conventional experimental tasks that require isolated and discrete cognitive processes (example, Stroop Task) report H values that lie closer to the “high experimental control” end of the spectrum.

Table 1. Hurst Values in Greymatter, DMN, and Visual Network.

Main pathways of regulation of local blood flow during neuractivation.

Paradigm design.

Figure 2. Overview of the system pipeline. In eye tracking (left column), the initial eye corner region for tracking is manually selected when the subject is ready to start. The eye corner and pupil tracking is based on DCF-CSR [3] and adaptive thresholding [4]. Our gaze interface will build an initial regression model based on standard 9 points calibration [2]. The gaze interface will convert gaze into visual content interaction information which will progressively update gaze regression model.

Gaze interaction interface overview. The interface converts gaze into a 3D ray to test collision with 3D object’s collider. Subjects see the visual object inside the collider in the 3D world. Dwell time records how long the ray continuously intersects with the collider. Linked events trigger associated 3D world events. Effectors and animators trigger particle effects and corresponding animation of visual objects while the dwell time increases.

Figure 1. LC localization using a predefined LC atlas and the TSE image of each subject. In T1w structural space, we use a criterion (Student’s t-test) to determine a coarse LC location. Then, either TSE intensities in the LC mask or an LC atlas is transformed to EPI functional space for a more precise localization (using trilinear interpolation). The result is one of three possible outcomes: 1) we localize the LC structure within M_1SD; 2) we localize the LC structure within M_2SD; 3) we localize the LC structure with the predefined LC atlas (i.e., without using any information from I_TSE).

Figure 3. (a) GLM to estimate and test the contributions of PDB and PR to LC BOLD activity (controlling for the variance due to the presence of stimuli). Trial-to-trial variabilities of PDB and PR (V_PDB and V_PR) were modeled as boxcar functions with the amplitude of each trial modulated by the pupil measurements. The boxcar functions were convolved with a canonical double-gamma hemodynamic response function before fitting into the GLM. (b) Group level statistical analysis in testing regression weights against zero. *p < 0.05; **p < 0.01.

Figure 1. The stimulating location of taVNS on auricular surface.

Figure 2: The Pittsburgh Sleep Quality Index(PSQI) score of group A and B before and after treatment.