Graph-Laplacian Features for Neural Waveform Classification

Abstract

Analysis of extracellular recordings of neural action potentials (known as spikes) is highly dependent upon the accuracy of neural waveform classification, commonly referred to as spike sorting. Feature extraction is an important stage of this process because it can limit the quality of clustering that is performed in the feature space. Principal components analysis (PCA) is the most commonly used feature extraction method employed for neural spike recordings. To improve upon PCA's feature extraction performance for neural spike sorting, we revisit the PCA procedure to analyze its weaknesses and describe an improved feature extraction method. This paper proposes a linear feature extraction technique that we call graph-Laplacian features, which simultaneously minimizes the graph Laplacian and maximizes variance. The algorithm's performance is compared with PCA and a wavelet-coefficient-based feature extraction algorithm on simulated single-electrode neural data. A cluster-quality metric is proposed to quantitatively measure the algorithm performance. The results show that the proposed algorithm produces more compact and well-separated clusters compared to the other approaches.

Spike sorting of heterogeneous neuron types by multimodality-weighted PCA and explicit robust variational Bayes

“…Details are found in (Ghanbari et al, 2011). As in the case of PCA, the original D -dimensional data set X = {x n } N n = 1 is reduced to a D ′-dimensional data set through the transformation Y = A T X, where A = {a d } D ′ d = 1 and a d is a D -dimensional vector.…”

“…We compare the performance of mPCA with that of PCA, an improved multimodality pick-up algorithm (mPICK) and Graph Laplacian features (GLF), which project a high-dimensional data onto a low-dimensional space while preserving the topological (i.e., clustering) structure of the original data (Belkin and Niyogi, 2003; He and Niyogi, 2004). GLF is a linear mapping, solves the difficulties arising from the non-linearity of Laplacian eigen maps in a model-based clustering (Chah et al, 2011), and exhibits an excellent performance in spike sorting (Ghanbari et al, 2011). However, the computational cost of GLF increases drastically for larger data size.…”

Using density invariant graph Laplacian to resolve unobservable parameters for three-dimensional optical bio-imaging

“…The work by Raheleh Kafieh et al [7] and Hang Su et al [12] consider only segmentation of a single 2D microscopic image, and the work by Yasser Ghanbari et al [4] considers only time-domain analysis, and the work by A. Singer and H.-T. Wu [11] considers only 2D tomography for large number of sampling on a phantom model.…”

“…This graph-based algorithm has advantages of its simplicity, interpretability and provides efficient representation for complex data structure [1][2][3]. It has been used for bio-signal and bio-image processing such as spike sorting of neural signal [4], protein function prediction [5], image classification of MRI [6] and retinal layered image segmentation [7]. The application on 3D reconstruction of bioimaging is first addressed by Ronald R. Coifman et al [8,9] for cryo-electron microscopy (Cryo-EM), in which hundreds to thousands unknown orientations of view of a molecule could be ordered so that reconstructing 3D structure of a single molecule is possible.…”

Automatic spike sorting by unsupervised clustering with diffusion maps and silhouettes

“…see [1][2][3][4][5]. Quiroga [6] proposed a technique that combines wavelet transformation (WT) with superparamagnetic clustering without assumptions such as low variance or Gaussian distributions.…”