-- you have earned Featured Contributor Badge
Your exceptional post has been selected for our editorial column Staff Picks http://wolfr.am/StaffPicks and Your Profile is now distinguished by a Featured Contributor Badge and is displayed on the Featured Contributor Board. Thank you!
Hi Anshul, this is wonderful post! Nice to see a medical application of machine learning.
BTW. I think besides CNN and RNN, another architecture worth trying is TCNN (temporal convolution neural network), which is usually used when dealing with temporal series data with causal links.
Hey, I'm glad that you liked it! TCNN would probablty be a better architechture to implement, especially when dealing with sequences or patterns of artifacts.
Hi Anshul! Did you have any chance tried TCNN? (That Updating Name was me on cloud glitch)
I don't have much cardiac knowledge, but EEG as a 1 dimensional (quasi-)periodic signal, maybe it fits in some traditional analysis methods (Fourier, wavelets, etc.), at least as a pre-neural-network feature preparation parts.
Thanks for sharing your exploration of artifact classification. I know other OpenBCI users would be interested in your post/tutorial if there was some extra info on how you setup your electrodes & collected the EEG data. Basically, the parts prior to Data Cleaning.
I primarily refered to sentdex's video on Brain Computer Interface w/ Python and OpenBCI for EEG data. There are 3 major steps before the Data Cleaning:1.Assembly of Kit, 2. Setup of GUI, 3.Data Extraction.
1.My kit consisted of the Cyton+Daisy board, Ultra Cortex Mark 4 headset and, Wi-fi Shield. There is neat explanation by OpenBCI team for a step by step assembly procedure.
2.You can download the Opeb BCI GUI from here. Before the EEG data can be tranmitted, you'll have to select the following options on the GUI: LIVE & serial transfer protocol from System Control Panel, 16 channel count.
3.You can extract data right from the GUI into a python file by selecting ther Networking option and using the LSL (Lab Streaming Layer). Within the LSL layer, you can choose to extract Time Series, FFT or BandPower. Make sure you've selected the right number of floats for that channel .
Time Series- 1float/ch; FFT-128 float/ch; Band Power-5float/ch
And there you have it! Your data will be ready according to your specified sample rate which you can copy-paste onto a Wolfram notebook.