Mechanisms of individualized fMRI neuromodulation for visual perception and visual imagery

Neuromodulation is a growing precision-medicine approach to modulating neural activity that can be used to treat neuropsychiatric, and general pathophysiologic conditions. We developed individualized fMRI neuromodulation (iNM) to study the mechanisms of visuospatial perception modulation with the long-term goal of applying it in low-vision patient populations having cortical blindness or visuospatial impairment preceding subjective cognitive impairment. To determine these mechanisms, we developed a direction and coherence discrimination task to engage visual perception (VP), visual imagery (VI), selective extero-intero-ceptive attention (SEIA), and motor planning (MP) networks. Participants discriminated up and down direction, at full and subthreshold coherence under iNM or control (no iNM). We determined the blood-oxygen-level-dependent (BOLD) magnitude as the area under the curve (AUC) for VI, SEIA, and MP encoded networks and used a decoder to predict the stimulus from brain maps.

Methods
Subjects

Eight healthy, right-handed volunteers (4 males, 4 females, age range = 25-31) were recruited into this 3-day study after obtaining informed consent according to the Baylor College of Medicine Institutional Research Board. Exclusion criteria included prior and current medical or psychiatric diagnoses, intake of any medications, and general contraindications against MRI examinations. Participants had normal or corrected-to-normal visual acuity with MRI-compatible glasses. At the end of each study day, participants were compensated for their time.

MRI and fMRI Pulse Sequence Parameters

Structural and functional brain imaging was performed at the Core for Advanced Magnetic Resonance Imaging, at Baylor College of Medicine, Houston, Texas using a 3.0 T Siemens Prisma (Siemens, Erlangen, Germany). We used a 20-channel head/neck receiver-array coil to acquire images. A T1-weighted 3D magnetization-prepared, gradient-echo (MPRAGE) sequence acquired 192 high-resolution axial slices [field-of-view (FOV) = 245 x 245 mm²; base resolution = 256 x 256; repetition time (TR) = 1,200 ms; echo time (TE) = 2.66 ms; flip angle (FA) = 12°]. Functional data consisted of 33 interleaved axial slices acquired using an Echo Planar Imaging (EPI) sequence (FOV = 200 x 200mm², voxel size = 3.1 x 3.1 x 3.0 mm, TR = 2000 ms; flip angle = 90°, number of volumes = 244).

Real-time fMRI Neuromodulation Acquisition

Turbo-BrainVoyager (TBV; 2.0; Brain Innovation, Maastricht, The Netherlands) software was used to perform the following five pre-processing computations on EPI images acquired at every time repetition (TR): 1) 3D motion correction; 2) incremental linear detrending to remove BOLD signal drifts; 3) statistical brain map displays generated from a general linear model (GLM) along with beta weights (BOLD signal intensity values) for each condition; 4) extraction of average BOLD signal intensity values from individualized networks acquired on Day 1 scans (see Data Analysis); and 5) presentation of the network average BOLD signal intensity via the neuromodulation interface (Figure 1A). The iRTfMRI neuromodulation (iNM) interface steps are summarized in Figure 1. To increase the signal-to-noise ratio (SNR), we used an exponential moving average (EMA) algorithm to high-pass filter the ROI BOLD average and suppress low-frequency noise components such as scanner drifts and physiological noise effects (e.g., heart rate and respiration). The EMA output was then low-pass filtered via a Kalman filter to eliminate high-frequency noise.

Task Design

A random dot kinematogram (RDK) was presented to the lower right quadrant of each subject’s right visual field, while they were asked to fixate on a dot in the middle of the screen. The RDK displayed upward or downward motion at either fully coherent or subthreshold levels. Four levels of coherent motion were chosen for this study; 100%, 84%, 66%, and 33%. Here we focus on the full and subthreshold coherence levels represented as C100 and C33 throughout this paper. Using their central vision to fixate on a dot in the middle of the screen, participants were asked to track the direction of RDK motion, which was presented in the lower quadrant of their right visual field, through their peripheral vision as it alternated between directions versus random motion. In the control and neuromodulation conditions, participants were asked to superimpose the upward or downward direction of motion centrally via visual imagery, while the direction of motion was tracked via their peripheral vision. In the iNM condition, the central dot served as the neuromodulation interface, i.e., when the central dot filled with red color, it corresponded to the successful visual imagery of the upward or downward direction of motion. The direction of motion was interleaved with blocks of random motion, during which subjects were asked to rest by disengaging from superimposing imagery of the direction of motion while continuing to fixate on the central dot.

Study Structure

Our study included two sessions; each consisting of ten functional (echo planar imaging; EPI) scans, which included five control-no iNM scans that alternated with five neuromodulation (iNM) scans. Each EPI scan included eight continuous periods each lasting 8 minutes and 12 seconds. Within each period, subjects were cued to imagine motion perception as either up or down depending on the RDK session displaying one of the four coherence levels. Each coherent motion block lasted 20 secs and was interleaved with a baseline-random motion block (10 secs). The direction of coherent motion blocks was randomly counterbalanced across runs, following three rules: 1) each coherent motion block occurred twice during each period; 2) a coherent motion block was never followed by the same coherence level; and 3) each run consisted of a unique block order.

Neuromodulation Paradigm

Neuromodulation was determined by the color and extent of a circle that was filled, representing the magnitude and extent of each subject’s targeted network. The neuromodulation signal was calculated by comparing the percent BOLD signal change (PSC) generated during each control run for each coherence block with the rest block that preceded it. The BOLD PSC change was calculated from each participant’s individualized areas every 2 seconds as follows: BOLD PSCi(j) = 100% * [ROIs BOLD during Up OR Down direction selectivity i(j) –ROIs BOLD during tongue at rest i(j)]tongue rest i(j) where i represents coherence level (C100; C84; C66; C33), j represents the time interval (2 secs) used to compute the BOLD PSC of coherence level I . The neuromodulation presented at each TR was computed by comparing the current PSC value with a PSC reference range that included seven bins of 25% BOLD increase or decrease: -100%; -75%; -50%; -25%; 0; 25%; 50%; 75%; 100%. During the iNM run following each control run, if the PSC at a given time point was within the reference range or higher than the maximum value, the circle was filled with red, representing upregulation of the targeted ROI BOLD signal that controlled visual perception and imagery. If the PSC was lower than the minimal value, the circle was filled with blue, representing the downregulation of the targeted ROI BOLD signal that controlled visual perception and imagery.