Extraction of Neural Activity from In Vivo Optical Recordings Using Multiple Independent Component Analysis

Abstract

In vivo optical recording using voltage-sensitive dyes enables neural activities in the cortex to be visualized as two-dimensional images. However, the signals provided are often contaminated with strong physiological noise, which mainly comes from pulsation and respiration. An effective approach is independent component analysis (ICA) to extract the neural activity component from these data without averaging. It has been used to denoise data obtained optical recordings.ICA extraction has problems in three points: 1) the appropriateness of ICA solutions is not guaranteed a priori; 2) in most ICA models, there is linear mixing of mutually independent random variables although the mixtures are not random variables; 3) considerable computational effort is often required because of repetitive large-matrix calculations.We have developed a multiple ICA method that resolves these problems and promises more efficient extraction of neural activity components. Multiple ICA overcomes the above problems by performing segmented and repeated ICAs and by using an algorithm based on the time structure of the variables. Multiple ICA repetitively performs a small ICA for a segmented region consisting of several adjacent channels. By sliding a segmental template channel by channel over the whole measurement region ( Fig. 1), we can obtain multiple solutions for each channel. The most appropriate solution is uniquely selected from these multiple solutions based on a predetermined criterion.We confirmed the effectiveness of multiple ICA by comparing its performance with that of conventional (batch) ICA, particularly in terms of the strength of the extracted neural activities in guinea pig auditory cortex evoked by click sound stimuli. The pulsatile component was reduced, and the neural activity was revealed by both batch and multiple ICA (Fig. 2). Multiple ICA extraction bears comparison with batch ICA extraction, although the computational cost of batch ICA is higher than that of multiple ICA.We found that this multiple ICA method would be a powerful way to extract neural activity from large-scale optical recordings. Non-memberA multiple independent component analysis (ICA) method based on the noisy time-delayed decorrelation algorithm is described that overcomes the problems and improves the usefulness of conventional ICA, which is commonly used for extracting the actual neural activity from data measured using optical recording with a voltage-sensitive dye to visualize neural activities in cortical areas as two-dimensional images. The problems with conventional ICA extraction include the lack of an a priori guarantee that the solution will be appropriate, the linear mixing of mutually independent random variables although the mixtures are not random variables but time signals in many applications, and the general requirement for repetitive calculation of large matrices. Application of multiple ICA to the extraction of neural activities in the guinea pig auditory cortex evoked by both click sounds and pure tones from optical rec...