Xenon LFP Analysis Platform Is a Novel Graphical User Interface for Analysis of Local Field Potential From Large-Scale MEA Recordings

Abstract

High-density multi-electrode array (HD-MEA) has enabled neuronal measurements at high spatial resolution to record local field potentials (LFP), extracellular action potentials, and network-wide extracellular recording on an extended spatial scale. While we have advanced recording systems with over 4,000 electrodes capable of recording data at over 20 kHz, it still presents computational challenges to handle, process, extract, and view information from these large recordings. We have created a computational method, and an open-source toolkit built in Python, rendered on a web browser using Plotly's Dash for extracting and viewing the data and creating interactive visualization. In addition to extracting and viewing entire or small chunks of data sampled at lower or higher frequencies, respectively, it provides a framework to collect user inputs, analyze channel groups, generate raster plots, view quick summary measures for LFP activity, detect and isolate noise channels, and generate plots and visualization in both time and frequency domain. Incorporated into our Graphical User Interface (GUI), we also created a novel seizure detection method, which can be used to detect the onset of seizures in all or a selected group of channels and provide the following measures of seizures: distance, duration, and propagation across the region of interest. We demonstrate the utility of this toolkit, using datasets collected from an HD-MEA device comprising of 4,096 recording electrodes. For the current analysis, we demonstrate the toolkit and methods with a low sampling frequency dataset (300 Hz) and a group of approximately 400 channels. Using this toolkit, we present novel data demonstrating increased seizure propagation speed from brain slices of Scn1aHet mice compared to littermate controls. While there have been advances in HD-MEA recording systems with high spatial and temporal resolution, limited tools are available for researchers to view and process these big datasets. We now provide a user-friendly toolkit to analyze LFP activity obtained from large-scale MEA recordings with translatable applications to EEG recordings and demonstrate the utility of this new graphic user interface with novel biological findings.

Keywords: HD-MEA; LFP analysis; Plotly Dash; SCN1a; seizures.

FIGURE 1

High-density multi-electrode array (HD-MEA) data-analysis pipeline. The data processing for LFP activity detection and network analysis starts by selecting a group of about 600 channels that overlay the brain slice or region of interest, which are exported from the original hdf5 measurement raw file (4,096 channels, sampled at 10 kHz) to a reduced hdf5 file. This reduced file is further downsampled from 10 kHz to a desired frequency. This will be the working hdf5 file for the Xenon LFP Analysis Platform. (3Brain Logo: ^©Copyright 3Brain AG, Python Logo: ^©Copyright Python Software Foundation, Plotly’s Dash Logo: ^©Copyright Plotly).

FIGURE 2

Snapshot of the analysis GUI features. A view of the analysis GUI which is rendered in an html browser built in Python using Plotly’s Dash. The GUI has several interactive features from individual and group channel selection, low-pass, high-pass, and band-pass filtering, viewing entire trace or a small section of the trace, Fast Fourier Transformation (FFT) of sections from selected traces, customized raster plots, small groups of channels, and generation of group summary measures.

FIGURE 3

Example visualizations generated from the GUI including raster plots, time-series traces, LFP activity peaks, and time frequency transformations. (A) 407 channels selected for this analysis representing the MEA sensor spatial array, covering both the neocortex and the hippocampal regions. (B) Raster plot for all the channels in the working file irrespective of brain regions for a selected time range, demonstrating time points of when the activity occurs in the slice. (C) Three selected traces: the blue trace from the hippocampal region, the red trace from one end of the neocortex, and the aqua trace from the other end of the neocortex. (D) Zoomed in view of the first seizure [the region bracketed in panel (C) as AA], with the black dashes showing a peak find function within the GUI. (E) This demonstrates the ability to plot filtered traces along with the raw traces. (F) The amplitude frequency transformations for the traces in panels (D,E) (band-pass filtered). The filtered FFT spectrum for each is shown in purple.

FIGURE 4

Channel groups and raster plot can be generated to visualize LFP activity in different regions of the brain slice. (A) Sensor locations corresponding to three different regions selected for analysis and the region-specific raster plots. Group 1 being the hippocampus, while groups 2 and 3 each being one half of the Neocortex. (B) Summary plots and measures that can be generated within the analysis platform.

FIGURE 5

Simple and fast unsupervised seizure detection method. (A) The raw trace from the recording downsampled to 300 Hz frequency for a sample channel. (B) Spectral activity calculated from the Short-time Fourier Transform using Hanning Window, for a time window of 1 s with no overlap. (C) LFP activity is detected using a threshold of 6 standard deviations from the baseline voltage for each individual channel and a fixed duration of 0.035 s, followed by applying two sets of sliding windows (length 30 datapoints and 500 datapoints) to detect time regions of continuous activity. (D) Spectral activity is detected when the magnitude is greater than mean + 6 standard deviations from the baseline spectrum magnitude. The spectral activity is also passed through two sliding windows to detect regions of continuous spectral activity. (E) The overlapping regions of LFP activity and spectral activity of 10 s or more are used to identify the seizure envelop. The start of seizure is primarily identified using the time point when the spectral activity is greater than 6 standard deviations from the baseline.

FIGURE 6

Seizure activity tracking over space and time. (A) Seizures in individual channels in a group are automatically detected. Their respective start and end times can be tracked across channels in that group. (B) Regions of the raster between time intervals can be selected as demonstrated to generate seizure maps of selected brain regions within the interval. (C) Seizure map for the time interval selected and channels in the group, including initiation site of the seizure, maximum distance the seizure spread from the initiation point, duration of the seizure, and the rate of seizure spread across the tissue.

FIGURE 7

Scn1aHet mice have an altered seizure pattern in the low Mg²⁺ model. (A) Example raster plots from a control brain slice and a brain slice from Scn1aHet mice. (B) Scn1aHet mice have significantly more SLE than littermate controls (Mann–Whitney test, p = 0.04, n = 7–9 slices). (C) From the brain slices that displayed SLE, Scn1aHet demonstrated a significant increase in time to first seizure compared to littermate controls (unpaired t-test, p = 0.006, n = 5–7 slices). (D) SLE from the Scn1aHet propagate significantly faster than seizures in the littermate controls (Mann–Whitney test, p = 0.0059, n = 10–14 seizures from 5 to 7 slices). The first two seizure from the slices that had SLE were used in this analysis. (E) Seizure duration was not different between Scn1aHet and littermate controls (Mann–Whitney test, p = 0.66, n = 10–14 seizures from 5 to 7 slices). The first two seizures from the slices that had SLE were used in this analysis. *P < 0.05, **P < 0.01.