Machine Learning Approaches to Classify Anatomical Regions in Rodent Brain from High Density Recordings.

Identifying different functional regions during a brain surgery is a challenging task usually performed by highly specialized neurophysiologists. Progress in this field may be used to improve in situ brain navigation and will serve as an important building block to minimize the number of animals in preclinical brain research required by properly positioning implants intraoperatively. The study at hand aims to correlate recorded extracellular signals with the volume of origin by deep learning methods. Our work establishes connections between the position in the brain and recorded high-density neural signals. This was achieved by evaluating the performance of BLSTM, BGRU, QRNN and CNN neural network architectures on multisite electrophysiological data sets. All networks were able to successfully distinguish cortical and thalamic brain regions according to their respective neural signals. The BGRU provides the best results with an accuracy of 88.6 % and demonstrates that this classification task might be solved in higher detail while minimizing complex preprocessing steps.

When the Ostrich-Algorithm Fails: Blanking Method Affects Spike Train Statistics.

Modern electroceuticals are bound to employ the usage of electrical high frequency (130-180 Hz) stimulation carried out under closed loop control, most prominent in the case of movement disorders. However, particular challenges are faced when electrical recordings of neuronal tissue are carried out during high frequency electrical stimulation, both in-vivo and in-vitro. This stimulation produces undesired artifacts and can render the recorded signal only partially useful. The extent of these artifacts is often reduced by temporarily grounding the recording input during stimulation pulses. In the following study, we quantify the effects of this method, "blanking," on the spike count and spike train statistics. Starting from a theoretical standpoint, we calculate a loss in the absolute number of action potentials, depending on: width of the blanking window, frequency of stimulation, and intrinsic neuronal activity. These calculations were then corroborated by actual high signal to noise ratio (SNR) single cell recordings. We state that, for clinically relevant frequencies of 130 Hz (used for movement disorders) and realistic blanking windows of 2 ms, up to 27% of actual existing spikes are lost. We strongly advice cautioned use of the blanking method when spike rate quantification is attempted. IMPACT STATEMENT: Blanking (artifact removal by temporarily grounding input), depending on recording parameters, can lead to significant spike loss. Very careful use of blanking circuits is advised.

Automated drill-stop by SVM classified audible signals.

Neuroscience research often requires direct access to brain tissue in animal models which clearly requires opening of the protective cranium. Minimizing animal numbers requests only well-experienced surgeons, since clumsy performance may lead to premature death of the animal. To minimise those traumatic outcomes, an algorithmic approach for closed-loop control of our Spherical Assistant for Stereotaxic Surgery (SASSU) was designed. Controlling the surgical robot's micro-drill unit by audio pattern recognition proved to be a simple and reliable way to automatically stop the automated drill feed. Sound analysis based on the anatomical morphology of a rat skull was used to train a Support Vector Machine (SVM) classification of the time-frequency representations of the drill sound. Fully automated high throughput animal surgeries are the goal of this approach.