Portal de Periódicos da CAPES

Lossless Compression of Neural Signals with Predictor Schemes Achieving more than fivefold data Reduction of in-vitro Recorded Retinal Signals

2018; Frontiers Media; Volume: 12; Linguagem: Inglês

10.3389/conf.fncel.2018.38.00034

1662-5102

Matteo Pagin, Florian Jetter, Günther Zeck, Maurits Ortmanns,

Analog and Mixed-Signal Circuit Design

Event Abstract Back to Event Lossless Compression of Neural Signals with Predictor Schemes Achieving more than fivefold data Reduction of in-vitro Recorded Retinal Signals Matteo Pagin1*, Florian Jetter2, Günther Zeck2 and Maurits Ortmanns1 1 Universität Ulm, Institut für Mikroelektronik, Germany 2 Natural and Medical Sciences Institute, Germany Motivation: State-of-the art CMOS-based in-vitro microelectrode arrays feature thousands of recording channels. In combination with the recording bandwidth required for action potential identification they generate of huge amount of data. A system employing 65536 channels [1] at a sampling rate of 10kHz and 12 bit resolution generates about 983Mbyte/s data rate. Similar data rates are obtained for the CMOS based MEA used here which employs 4225 channels sampled at 25 kHz at a resolution of 14 bit [2].Storage of such a huge amount of data can greatly benefit from lossless compression which reduces the data without any loss of information. A predictor based scheme was proposed in [3] and successfully used to compress pre-recorded data from neural implants. In this abstract the scheme is extensively tested on in-vitro retinal signals recorded from 4225 channels [2]. Material and Methods: 1) Predictor based compression: This scheme, also known as predictive or differential encoding, consists in using a predictor block to forecast the incoming sample of a signal using past information about that signal. After prediction the error is calculated by subtracting the true signal value and the predicted one. The distribution of the error presents a lower dynamic range and a more skewed distribution than the original signal. These conditions can then be exploited by an entropy encoder to effectively compress the signal (Figure 1). The compressed error is stored along with the predictor and can be used later for signal reconstruction. The compression is hence lossless since it is possible, using the saved predictor, to reconstruct exactly the original signal. 2) Predictor implementation: In this work a one-layer linear neural network is used to predict the incoming signal [3]. The neural network consists then of one neuron which makes its prediction by multiplying past values of the signal by a learned set of weights and summing them together. The weights are learned from a short segment of neural signal (~ 2000 samples) using a Levenberg–Marquardt algorithm which minimizes the least mean squared error between the original signal and the predicted value. The past values of the signal can be either samples coming from the same channel to be predicted or from adjacent channels, thus allowing the neural network to exploit spatial and temporal redundancy present in the recording. 3) Evaluation Metric: To assess the performance of the compression algorithm the measure of compress ratio is used here and defined as the number of bits required before the compression divided by the number of bits used after compression. Since compression is lossless there is no need to quantify information losses. Results and Discussion: The compression technique is extensively tested on a CMOS-based MEA of 65x65 channels of in-vitro recorded retinal ganglion cell spikes. Compress ratios for each channel is shown in Figure 2. An average data reduction of 5.6 is achieved with a standard deviation of approximately 0.28; demonstrating that the compression can effectively reduce the data rate and is overall quite stable over the whole array. The channels of one row (#57) appeared to be broken in the recording. It is possible to see that here the compression is worse than the average. Using spatial information combined with temporal information in this dataset does not provide an advantage resulting in an average compression of 5.58, which is only slightly worse than compression per single channel. Figure 1 Figure 2 References [1] David Tsai, Daniel Sawyer, Adrian Bradd, Rafael Yuste & Kenneth L. Shepard. A very large-scale microelectrode array for cellular- resolution electrophysiology. Nature Communications 8, 1802 (2017) [2] Bertotti G., Velychko D. et al., A CMOS-based sensor array for in-vitro neural tissue interfacing with 4225 recording sites and 1024 stimulation sites, 2014 IEEE Biomedical Circuits and Systems Conference (BioCAS) [3] Pagin, M.; Ortmanns, M., A Neural Data Lossless Compression Scheme Based on Spatial and Temporal Prediction, 2017 IEEE Biomedical Circuits and Systems Conference (BioCAS) Keywords: Neural signal compression, predictive coding, mea array compression, data recuction, neural signal processing Conference: MEA Meeting 2018 | 11th International Meeting on Substrate Integrated Microelectrode Arrays, Reutlingen, Germany, 4 Jul - 6 Jul, 2018. Presentation Type: Poster Presentation Topic: Microelectrode Array Technology Citation: Pagin M, Jetter F, Zeck G and Ortmanns M (2019). Lossless Compression of Neural Signals with Predictor Schemes Achieving more than fivefold data Reduction of in-vitro Recorded Retinal Signals. Conference Abstract: MEA Meeting 2018 | 11th International Meeting on Substrate Integrated Microelectrode Arrays. doi: 10.3389/conf.fncel.2018.38.00034 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 16 Mar 2018; Published Online: 17 Jan 2019. * Correspondence: Mr. Matteo Pagin, Universität Ulm, Institut für Mikroelektronik, Ulm, Germany, matteo.pagin@uni-ulm.de Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Matteo Pagin Florian Jetter Günther Zeck Maurits Ortmanns Google Matteo Pagin Florian Jetter Günther Zeck Maurits Ortmanns Google Scholar Matteo Pagin Florian Jetter Günther Zeck Maurits Ortmanns PubMed Matteo Pagin Florian Jetter Günther Zeck Maurits Ortmanns Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.