Performance evaluation of light source noise reduction algorithm based on independent component analysis for optical intrinsic signal data

Intrinsic optical signal (OIS) imaging technique is widely used in neuroscience research because it permits high resolution brain mapping without introducing molecular probes into the brain. However, low signal-to-noise ratio (S/N) of OIS is a serious problem that has to be resolved. So far, many algorithms have been developed to improve S/N. However, most of them require repeated acquisition of stimulus-to-response data and are therefore not suitable for OIS that express spontaneous activity of the brain. To overcome this problem, we developed an independent component analysis (ICA)-based algorithm for reduction of light source noise from OIS. The algorithm is based on a model of mixing mechanism of light source noise. It automatically determines the number of independent components and finds the component corresponding to light source noise based on similarity of power spectral densities. The noise component is removed by projecting the measured OIS onto the subspace orthogonal to the subspace spanned by the estimated noise component. Although usability of the algorithm was demonstrated by applying to real OIS data, some parameters were not optimized and quantitative performance was not clarified. In this study, we evaluated the noise reduction ability of the system by conducting performance test using synthetic OIS data containing light source noise. First, we identified the optimal parameter value for binning processing, which is applied prior to the noise reduction algorithm, based on accuracy of estimation of noise component. Second, we showed that reduction of light source noise by 10-20 dB was achieved under optimal conditions. These results indicate superiority of the algorithm and suggest its usefulness in improving S/N of real OIS data expressing spontaneous activity of the brain.