Aggie Szymanska – UCI Neuroinformatics


Projects

Action Potential (AP) Detection

Detection is a crucial component of any neurological signal processing algorithm. Although spike detection in general has been an issue for many years, the problem is not completely resolved. Neurological data is generally very noisy, therefore signal detection tools from other fields don’t readily apply. The noise coming from the recording hardware and environment is in most cases white Gaussian noise and is usually relatively easy to process. However, the presence of biological noise, including the activity of background neurons and ion-channel noise, is much more problematic to AP detection, because it is both correlated and statistically similar to the AP signal.

To tackle this issue we have used a machine learning approach to develop a multi-sensor matched filter for detecting multi-neuron extracellular APs, as well as a matched filter for multi-neuron calcium event detection (MMiCE) from fMCI imaged neuron populations.

Extracellular Action Potential (EAP) Detection

Multi-neuron Calcium Event Detection (MMiCE Detector)

 

Extracellular Action Potential (EAP) Classification

Multi-sensor extracellular recording is one of the most commonly used techniques for studying neural activity in vivo. Its main advantage is allowing action potentials (APs) to be recorded and resolved for multiple neurons simultaneously. However, in order to determine functional relationships and neural interactions, the recorded APs must be classified. Many techniques have been proposed for this purpose, ranging from principal component analysis (PCA), to expectation-maximization based clustering. Although these methods differ in their actual classification algorithms, the defining feature that sets most classification schemes apart is their feature selection algorithm. Choosing the correct feature largely determines the efficacy of extracellular action potential (EAP) classification, and features that can reliably represent a single neuron, as well as remain invariant across trials and sensor positions are preferable. One such classification feature is neuron location.

Mulit-sensor EAP Classification

 

Cardiomyocyte Analysis

We’ve developed a software package, MaDEC,  that can detect, quantify, and then compare cardiomyocytes treated with different drugs. The software can analyze time series data from various imaging techniques, and was tested on voltage sensitive dye imaged cells. The detection process relies on a matched filter to extract depolarization events, even at SNR levels as low as 3. The beating cardiomyocyte depolarization events (DEs) are then assessed across drug treatment, but extracting and comparing relevant DE features. (A.F. Szymanska et al. (2016))