EEG - Science method

Dear Eduardo,

thank you for your suggestion, I'll read the article you indicated.

Best regards,

Pawel

S. Béatrice Marianne Ewalds-Kvist Thank you very much for your guidance and response.

Here are some references that might be helpful for using Natus EEG in ERP studies:

Aamir Sharif Many thanks for your assistance!

Data points are the recorded samples. EEG signal is mainly an analogue signal, but it's digitized by sampling this signal into points. Number of samples(points) recorded per second depends on the "Sampling Frequency", which is a characteristic of the equipment and represents how many readings it can give in a second. Increasing the sampling frequency surely increases the accuracy of analysis, but requires much more computational power to process this larger amount of data. All this occurs for each channel. Example: if you have an equipment that records through 8 channels with a sampling frequency of 256Hz, then each channel gives 256 (points, samples, values) per second. I hope you got me right.

Integrating multimodal neuroscience data with Gen AI models requires careful attention to preprocessing, feature extraction, and multimodal fusion techniques.

Advanced tools such as TensorFlow, PyTorch, GNNs, and transfer learning help bridge different data types, leading to more powerful and predictive AI models capable of synthesizing complex neural and behavioral patterns.

Stephen Shannon Muse is probably not a perfect product, but it is certainly cheap and easy to handle. Obviously, it cannot be considered as effective as a professional medical EEG, but it provides very interesting basic information. I use it with sportsmen, and for them I establish cognitive abilities before sporting events. It's all at an experimental level, so maybe in a year's time there will be something better (that's how technology works). You can also use it during competitions because you just wear a lycra band if it's Muse2, or MuseS is already an elastic band. There are many softwares you can use with Muse, one of which my team created especially for it: it provides real-time data or you can download the data in csv (it is a commercial product). The software produces insights created specifically for the sporting environment. There are alternatives to Muse: they are all more expensive. But they should be tried. If you need more info write to me at rosso.gianluca@gmail.com

I am afraid that is not ideal. Using a single electrode to record in vivo local field potentials (LFPs) in the hippocampus is generally not recommended because: 1. Lack of Reference Point: LFP recordings measure the summed electrical activity of populations of neurons. Without a separate reference electrode, it’s hard to distinguish between true neural activity and background noise or artifact. A single electrode does not provide a way to differentiate local changes in electrical potential from broader shifts in the electrical field. 2. Volume Conduction: The hippocampus is embedded in a complex neural network, and the electrical signals recorded by a single electrode can include not just local activity but also signals that are conducted from other brain regions. This "volume conduction" effect makes it difficult to isolate hippocampal activity with a single electrode. 3. Spatial Resolution: A single electrode does not provide information about spatial variations in the neural activity across different regions of the hippocampus. This is especially problematic because the hippocampus has distinct anatomical and functional regions (e.g., CA1, CA3, dentate gyrus) that can have different activity patterns. Multi-electrode arrays or tetrodes allow for simultaneous recordings from multiple locations, improving spatial resolution and allowing for more precise mapping of hippocampal activity. 4. Difficulty in Signal Interpretation: LFPs represent a mixture of signals from excitatory and inhibitory neurons, synaptic potentials, and even glial cell activity. With only a single electrode, it is much harder to interpret which sources are contributing to the recorded signal, making the analysis less reliable. 5. Signal-to-Noise Ratio (SNR): Single electrodes are prone to picking up noise, which can be difficult to filter out when there is no comparison or control signal from nearby regions. Using multiple electrodes allows for better noise cancellation and more accurate isolation of the signal of interest. In summary, using a single electrode for LFP recordings in the hippocampus is limited because of issues with reference point clarity, signal contamination from volume conduction, reduced spatial resolution, and difficulties in interpreting the resulting signal. Multi-electrode setups help overcome many of these issues, providing a more accurate representation of hippocampal activity.

-----------------

However, it's still technically possible if necessary for specific experimental constraints. Here is a draft protocol: 1. Preparation - Anesthetize the animal with an appropriate anesthetic (e.g., isoflurane or ketamine/xylazine) and place it in a stereotaxic frame to secure the head. Ensure that the animal is fully anesthetized by monitoring vital signs (e.g., absence of the pedal reflex). - Shave the scalp and sterilize the area with ethanol and iodine. Make a midline incision to expose the skull. - Use a stereotaxic atlas to determine the precise coordinates for the hippocampus. Mark the area on the skull. 2. Craniotomy and Electrode Placement - Carefully drill a small hole in the skull over the target location using a micro-drill. - Insert the reference electrode in a distant brain region (e.g., the cerebellum or a peripheral muscle). Fix the ground/reference electrode securely. - Lower the recording electrode: Using the stereotaxic manipulator, slowly lower the single recording electrode into the hippocampus. Advance the electrode slowly to avoid damaging the tissue. 3. Signal Amplification and Recording - Attach the recording electrode to a pre-amplifier and connect the reference electrode to the ground of the system. - Adjust filtering parameters: Set up the amplifier to filter the signals between 1 Hz and 300 Hz (the typical frequency range of LFPs). Adjust the gain of the amplifier to ensure the signal is visible but not saturated. - Monitor the signal: Observe the incoming LFP signal in real-time on the data acquisition system. Adjust the electrode depth if necessary to optimize the quality of the signal. LFPs are typically visible as slow oscillations in the frequency range of 0.5 to 100 Hz, depending on brain state. 4. Recording - Once the electrode is properly placed, begin recording LFP data. Record for a suitable time period depending on your experimental question (e.g., several minutes to hours). - Monitor physiological parameters: Ensure the animal remains properly anesthetized throughout the procedure and check for any signs of discomfort or distress. Continuously monitor vital signs. 5. Post-recording - Electrode removal: Once recording is complete, carefully remove the electrode without damaging the tissue. - Close the craniotomy: Clean the exposed area of the skull and apply dental cement to cover the craniotomy. Use sutures to close the scalp. - Postoperative care: Administer analgesics (e.g., buprenorphine) and monitor the animal until it recovers from anesthesia. Place the animal in a warm recovery area. 6. Data Analysis - Filtering: Apply additional digital filtering if necessary to isolate LFPs from noise. - Artifact removal: Identify and remove artifacts from movement, heartbeat, or respiration. - Spectral analysis: Perform frequency-domain analysis (e.g., power spectral density) to examine oscillations like theta (4-8 Hz), gamma (30-80 Hz), or sharp wave-ripples (~100-200 Hz). - Signal-to-noise analysis: Evaluate the quality of the recording and signal-to-noise ratio.

Any suggestions?

Hello,

I think you should insert time triggers at the beginning of every trial. This way, you could select the data from the period you want.

Subcortical evoking potentials of the limbic, motoric, brainstem triad Dr. Darryl Pokea drdarrylpokea.com

what do you mean by active patients, and how many they are?

Dear Ola Badarani- Zahalka

I will endeavor to provide the best possible response, with the understanding that this will be of assistance to you. Furthermore, I hope that I have accurately interpreted your question.

1. When referencing EEG data to the common average, it is essential to exclude channels that are not being recorded, such as P7, P8, TP7, and TP8, if these are affected by the cochlear implant processor. The inclusion of non-functional or noisy channels can result in the skewing of the common average and the introduction of artifacts. It is therefore recommended that these channels be excluded from the re-referencing process.

2. Interpolation may be conducted at various stages, but it is typically advised to interpolate the channels after re-referencing and before ICA. This facilitates the assurance that the ICA decomposition is based on a comprehensive set of channels, which can result in a more precise identification of artifacts. The following represents a typical pipeline:

Preprocessing (filtering, bad channel removal)

Re-reference to the common average (excluding bad channels)

Interpolate bad channels

Run ICA to identify and remove artifacts

3. The ICA process is then applied to identify and remove artifacts. ICA can be run on epoched and baseline-corrected data, which may be preferable for specific types of analyses, particularly when dealing with event-related potentials (ERPs). However, ICA is traditionally applied to continuous raw data, as this allows the algorithm to have more data with which to work, thereby improving its ability to identify and separate components. Should one elect to run ICA on epoched data, it is imperative that the epochs be sufficiently lengthy and encompass a diverse array of data, thus ensuring accurate ICA decomposition.

I must profess a certain lack of familiarity with Hamming window filters, and thus I am hopeful that another individual may be able to provide insight on this matter.

Sincerely,

Florian

i am not aware

Strictly speaking, I think the delta band is 0-4 Hz. However, it's generally a good practice to eliminate very low frequencies due to the aforementioned issues. Although I have no direct experience with polysomnography data, considering what you mentioned about the article, your EEG data might be as expected. To decide whether to eliminate frequencies below 0.5 Hz, it could be helpful to review other articles to see what they usually report as important frequencies.

Hi Anisa,

Well, CSP is originally defined to filter the data in a way that results in two distinct groups. But, there are one-versus-one and one-versus-rest algorithms that make this method applicable to multiclass datasets.

Similarly, LDA was originally developed for binary classification but it can be extended to handle multiclass classification.

Fatemeh

Make both of them at 200Hz.

You can do that using EEGlab.

THANK YOU !

Hi Mojtaba Ameli

Would you please share the answer if you have found out how to interpret the threshold file? What is the difference between these thresholds?

I'm currently using Loreta software for connectivity analysis and i truly need some help with interpreting the result

You can try publishers with agreements and discounts with your institution to cover the APC

Recent SS posts in

- and in https://www.researchgate.net/post/What-criteria-do-you-have-in-mind-while-characterizing-a-scientific-theory/30 , pages 29, 30, though,

- are relevant to this thread question.

Cheers

Hi,

Theta activity likely signals transitions between events or behaviours. Complex spikes (occurring outside theta activity) may then trigger modifications to the efference copy (our internal model of how our actions affect the world).

Just my thought.

Hello Greatson. I’m assuming you have your data in long/tall format (i.e., one row for each observation), and that your baseline covariate is the first observation of your outcome variable. If I have that correct … then you do not need to control for the baseline covariate. This is handled with your time variable (as a within variable). To find differences across groups, duplicate your last line of code (i.e., the EEMEANS for the interaction) and modify it to “COMPARE(Cluster)” instead.

Hello,

To study the lexical-semantic system access in patients with cranial injury using fMRI and EEG, measure brain activity during language tasks and analyse the data for patterns related to lexical processing and semantic understanding.

Hope this helps.

Hello,

The Multimodal Developmental Neurology of Females with ASD dataset in NDA contains metadata, demographics, fMRI, and EEG data. Extract resting state EEG and fMRI via NDA's query tool.

Hope this helps.

First of all, I want to encourage you not to give up on an area just because there are a lot of researchers in it. People should follow their interests if they are capable of managing the task and are interested in them. It's not only that EEG research is a promising field, but it's also interesting to classify EEG data using machine learning or deep learning approaches. It's okay if it seems saturated to you. Improving already completed work is always a way to contribute. There are many ways to propose improved algorithms and models if you have an interest for mathematical modelling. Remember that even in well explored research fields, there is always space for creativity and advancement of interest.

It's better to start with a review paper on the latest research article in this field. In one paper (latest review paper), you can gain a clear idea of the work that has been done and the suggestions put forward by the authors (researchers) based on their investigation. This approach helps you understand the current state of the field and identify potential gaps or areas for further exploration.

In the biomedical field, preference should be given to applications that demonstrate effectiveness in promoting health and safety.

1. And, I would like to suggest that you integrate ML/DL techniques for EEG classification along with IoT or some real-time device, such as Jetson Nano or an equivalent.

2. EEG signals should have noise and limited spatial resolution. Maybe you can investigate.

3. Left and right hand movements generate distinct EEG signals. If you can collect a real dataset from reputable medical resources, you could investigate EEG signals in paralyzed individuals and analyze them.

I am sharing here some of the article maybe you can have a look, i feels that could help you better:

*) Current Status, Challenges, and Possible Solutions of EEG-Based Brain-Computer Interface: A Comprehensive Review.

*) A review on analysis of EEG signals.

*) Deep Learning Algorithm for Brain-Computer Interface.

*)Encyclopedia of Clinical Neuropsychology.

Finally, as this is your graduation thesis, it's important to have a backup plan. During research, numerous byproducts are often produced, many of which hold value. I hope you will successfully reach your final destination with this research. However, it's essential to keep proper track of your byproducts. They may prove invaluable in shaping your thesis and ensuring you graduate on time. Furthermore, even after graduation, consider continuing your research if possible.

You can import all the information from the data info. Yet you have to choose according to your file type, e.g., (edf,bdf), etc.

thank you so much sir.... I know that wavelet Decomposition breaks the signal like this that' s why I decided to opt for level 6 decomposition in order to get the delta frequency too, but the detail coefficients I am getting have very different dimensions so I just wat to sure ..that Are they OK

Look at this work: Sun S, Cao R, Rutishauser U, Yu R, Wang S. A uniform human multimodal dataset for emotion perception and judgment. Sci Data. 2023 Nov 7;10(1):773. doi: 10.1038/s41597-023-02693-z. PMID: 37935738; PMCID: PMC10630434.

You may find some useful information.

Have you tried the edfreader?

You can find it online or I can send it to you.

best

Will you please read what I have written in the near past !

Hello there, I am currently working on analyzing the EEG signal in the DEAP database. They have added [labels] columns to the EEG data; you can find them when downloading the entire dataset.

I am using Python. I used MATLAB also for the same reason, pleas if you have any questions do not hesitate to ask.

You can use the EEGlab toolbox on Matlab.

text me if you need any assistance.

best

Thanks. I believe this data will be helpful for my future research. How can I download the data?

I think that several artifacts contaminated these signals. On the other hand, what you are asking is not clear, you may inbox me to clarify the question better.

Yes, you are right, thanks for your helpful solution to this problem!

You can look for papers on this topic and check if the authors share their datasets in public repositories.

Hi,

Hope some of these references might be of help to you:

Single-Case Experimental Designs for Clinical Research and Neurorehabilitation Settings: Planning, Conduct, Analysis and Reporting

Robyn Tate; Michael Perdices, 2018

Franklin, Ronald D. Design and Analysis of Single-Case Research.2014

If I'm correct, a license is now required to access raw data, which wasn't the situation previously when access was free of charge. If you're using older versions of the Emotiv Epoc, you can continue to use the older version of OpenVIBE (0.11.0, available at http://openvibe.inria.fr/pub/bin/win32/) software in conjunction with the EmotivEpocInstaller_v1.0.0.4 SDK.

For additional details, please refer to this link: http://openvibe.inria.fr/how-to-connect-emotiv-epoc-with-openvibe/

Hi,

Your idea of using WE in EEG analysis for ADHD diagnosis seems feasible, especially if initial results show promise. Just ensure you validate it against traditional band-based methods.

Hope this helps.

run EEGLAB

in command line write:

M = csvread("path/file.csv")

then import the "M" array using the menu "File > Import data > From ASCII/float file or Matlab array"

Don't forget to enter the sampling rate and specify 'M' as a variable to import.

Thank you very much for the reply and the suggestion, after your answer i have revisited my Code and found an error during computation. Now everything works fine.

best regards

Yeah I echo Hsin-Yuan Chen's sentiments - if there are latency issues then switching to E-Prime won't fix them. In my (admittedly limited) experience with E-Prime, I run into far more than latency problems with E-Prime relative to PsychoPy.

Khaled Saad

Hi,

Please refer to the following articles,

Hope this helps. 

Many thanks for your answer. I am still working with GraphVar due to it has GUI. I have a question during the process.

I created the variable sheet and I encounter an error about file names while I execute it. The file name in the data~> CorrMatrix~> "CorrMatrix_P01EC1_Active" is the same as the variable sheet. I have attached screenshots of them. Actually, I tested both versions of the attached screenshots "Origin_of_errors1" and " Origin_of_errors2" and still have errors.

I would be really appreciated your help.

Thanks in a billion.

Neda

Not sure about goal but here's a step-by-step guide:

1. Set up your Python environment: Make sure you have Python installed on your system. You can download and install the latest version of Python from the official Python website (https://www.python.org/).

2. Create a new Python project: Set up a new directory for your Python package. This directory will contain the necessary files and folders for your package.

3. Define the package structure: Inside your project directory, create a folder with the name of your package (e.g., `nneten`). Inside the package folder, create the following files:

- `__init__.py`: This file makes the `nneten` folder a Python package.

- `nneten.py`: This file will contain the implementation of the NNetEn calculation.

4. Implement NNetEn calculation: Open `nneten.py` and define the necessary functions and classes to perform the NNetEn calculation based on the described methodology. This may involve importing relevant libraries, implementing entropy calculation algorithms, and any other required components.

5. Package documentation: Create a `README.md` file in your project directory to provide instructions and documentation for using your Python package. Include details about the NNetEn calculation method, usage examples, and any other relevant information.

6. Package installation: To make your package installable, create a `setup.py` file in your project directory. The `setup.py` file should contain information about your package, such as its name, version, author, dependencies, and other relevant details. You can refer to the official Python documentation on packaging for more information on creating a `setup.py` file.

7. Test your package: Create a separate directory for testing your package. Write appropriate unit tests to verify the correctness of your NNetEn calculation implementation. You can use the `unittest` module or any other testing framework of your choice.

8. Package distribution: Once you are satisfied with your package, you can distribute it on platforms like PyPI (Python Package Index) so that others can easily install and use it. Refer to the PyPI documentation for detailed instructions on how to upload and distribute your package.

Good luck

Hi,

There are several articles and sources on measuring customer satisfaction or brand loyalty through neuroimaging, specifically EEG. Here are some suggestions:

These sources provide insights into how EEG measures can be used to measure customer satisfaction or brand loyalty in neuromarketing research.

Hope this helps.

Hsin-Yuan Chen Thank you so much

reprocessing EEG data in CSV format typically involves several steps, such as data cleaning, filtering, and artifact removal. Here's a general outline of the preprocessing steps you can perform on EEG data in CSV format:

import pandas as pd # Load the EEG data from CSV eeg_data = pd.read_csv('eeg_data.csv')

from scipy.signal import butter, filtfilt # Butterworth bandpass filter function def butter_bandpass(lowcut, highcut, fs, order=5): nyquist = 0.5 * fs low = lowcut / nyquist high = highcut / nyquist b, a = butter(order, [low, high], btype='band') return b, a # Apply a bandpass filter to the EEG data def apply_bandpass_filter(data, lowcut, highcut, fs, order=5): b, a = butter_bandpass(lowcut, highcut, fs, order=order) filtered_data = filtfilt(b, a, data) return filtered_data # Define the cutoff frequencies for the bandpass filter lowcut = 1.0 # Lower cutoff frequency in Hz highcut = 30.0 # Upper cutoff frequency in Hz # Define the sampling frequency (adjust according to your data) fs = 250 # Sampling frequency in Hz # Apply the bandpass filter to the EEG data filtered_data = apply_bandpass_filter(eeg_data.values, lowcut, highcut, fs)

These are the general steps involved in preprocessing EEG data in CSV format. However, the specific preprocessing steps may vary depending on your research question, the characteristics of the EEG data, and the analysis techniques you plan to use. It's always important to consider the specific requirements of your study when preprocessing EEG data.

Hi,

This is a basic outline; more complex tasks may necessitate advanced methods and specialised toolboxes like EEGLAB or FieldTrip.

Hi,

EEG can be synchronised with smartphone triggers using Bluetooth, Wi-Fi, or a hardware device. Apps like LSL Trigger can be used to send triggers from social platforms to EEG recording software.

Try https://physionet.org/.

In extracting frequency bands from EEG signals, using PSD is a more suitable approach. PSD provides an accurate representation of the distribution of power across different frequencies, which is ideal for identifying alpha, theta, beta, and gamma bands. While the absolute value of the STFT can illustrate how the signal's spectrum changes over time, it may not adequately distinguish the power contribution of the different frequency bands. Thus, to isolate these bands, compute the PSD from the complex-valued STFT and segment the resulting spectrum into the respective frequency bands.

Hello dear professor.

I am currently a PhD student in physiology in Iran and I am accepting my thesis papers. I am interested in studying in the post-doctoral course. If you are familiar with the conditions of my country, we are facing severe sanctions and hard research work. I also barely finished the thesis. I need to get financial aid to be able to enter this course. And how good that your Ph.D. course is 3 years, Iran we finish this course in 4 years, but under the strict conditions of 5 years of experience. Your number one project is exciting.

Thank you very much

Use the eeglab function

eeg_regepochs

You may find the published EEG dataset of Alzheimer, Frontotemporal dementia and Control recordings here: https://openneuro.org/datasets/ds004504

And the Data Descriptor paper that accompanies this dataset here:

What makes this dataset truly invaluable is its potential for significant reuse in Alzheimer's EEG machine learning studies. With a lack of publicly available EEG datasets, researchers now have a unique opportunity to explore brain activity and connectivity alterations, develop new diagnostic approaches, and pave the way for innovative treatment methods.

Join us in unlocking the potential of EEG data to further our understanding of Alzheimer's disease and frontotemporal dementia. Let's collaborate, explore new frontiers, and work towards developing groundbreaking advancements in diagnosing and treating these conditions. Together, we can make a difference!

Aparna Sathya Murthy , Alireza Falakdin : Thank you so much for your answers.

Hi,

Try in GITHUB.

To download and extract a tar.gz file, you can follow these steps:

Note: The x flag is used to extract the files, z is used to decompress the gzipped file, and f is used to specify the filename.

I found that someone also wants to know the answer. Thus, I try to solve my own problem.

I have written an e-mail to Professor Veniero who published an article called "TMS combined with EEG: Recommendations and open issues for data collection and analysis". She thought it might be a bit more difficult to clean the data, but it is feasible.

I hope this answer can help others an if anybody wants to discuss the question with me you can leave a message or send e-mail to tianqingl@stumail.ysu.edu.cn

Bhogaraju Anand It is useful. Thank you for your answer.

You can simply divided them on left and right ones and also you can swich off the Fz, Cz, Pz and Oz. Do you need me to send you an example?

Hello dear,

It seems your question comes form one paper which determine the brain activity between frequencies and some tasks. I would say that

The statement that the power spectral density (PSD) of electroencephalogram (EEG) is inversely correlated with brain activity is incorrect. The EEG power spectral density measures the power of the EEG signal at different frequencies and reflects the distribution of electrical activity in the brain across different frequency bands. The PSD does not indicate the strength or intensity of brain activity as a whole, but rather the distribution of activity across different frequency bands.

Higher power in a particular frequency band within the EEG power spectrum indicates greater neural activity in that specific frequency range. For example, higher power in the alpha frequency band (8-12 Hz) may indicate greater cortical inhibition, while higher power in the beta frequency band (13-30 Hz) may indicate greater cortical activation.

In summary, EEG power spectral density is a measure of the distribution of electrical activity across different frequency bands, and higher power in a specific frequency range indicates greater neural activity in that frequency range, rather than an overall level of brain activity.

Let's discuss..

Hi,

Yes, it is the Time Series model I was referring to.

Jakob Sajovic thank you so much for your thorough and extremely helpful response.