Calcium Imaging Data Analysis

We present a novel parameter-free independent component analysis (ICA)-based method, ICA with signal reconstruction and ordering (ICA+SRO), to automatically extract spatial and temporal neuronal signals from calcium imaging sequences. The method, originally proposed in [1], begins by applying a median filter and z-scoring each frame in the video in order to remove the majority of the high frequency noise. Next, PCA (principal component analysis) then ICA are performed. Then, the sign of the resulting components and time courses is adjusted according to the sign of the skewness of the component, thus removing the sign ambiguity inherent to ICA. This is followed by a binarization step, where the threshold is set to be pure-white, and the largest object in the binary image is selected. This step removes the majority of pixels not associated with neuronal activation. Next, for each component, the average pixel intensity of the thresholded object for each frame is computed, producing the final timecourse for that component. This step resolves the scaling ambiguity inherent to ICA. In the final step, the components are sorted based upon the skewness of their timecourses, which automatically sorts the components from most likely to be true neuronal signals to those components that are least likely, thus simplifying any further studies.

    Independent component analysis with signal reconstruction and ordering (ICA+SRO) (ICA+SRO) [1]


[1] Y. Levin-Schwartz, D. R. Sparta, J. F. Cheer, and T. Adali, "Parameter-free automated extraction of neuronal signals from calcium imaging data," In the proceedings of 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), New Orleans, LA, 2017, pp. 1033-1037.