Fig. 2

Schematic of the processing algorithms. a Schematic of the algorithm for large artefacts detection. The dimensionality of SLS data is reduced using PCA and the PC1 scores vector is fitted using an autoregressive model of second order (AR(2)). If the sliding correlation (computed in a 10 s window) between the PC1 scores vector and its fit drops below a given threshold (set to 0.3), a large artefact is detected (red star). b Left: an SLS trajectory (yellow line) with a surround of 4 pixels is overlapped to a projection of the corresponding raster acquisition and its reference segmentation (red ROIs). A reference box (yellow square) is scanned at the end of the SLS trajectory. Right top. Pixels assignment for an example ROI (red box in left panel). Pixels can be assigned to the ROI (red dots), to its outer ring (yellow dots), to its surround (green dots), or to background (black dots). Right bottom. Pixels assignment for an example ROI of a SLS trajectory without ROIs surround. c Schematic of the algorithm for background activity subtraction. A 1-rank representation of the fluorescence activity of trajectory pixels labelled as background (black pixels) is computed using PCA. The across-pixels averaged low-rank representation is considered as a proxy of background activity, multiplied by 0.7 and subtracted from the fluorescence of all the pixels. d Schematic of the small and local artefacts correction algorithm based on NoRMCorre. Activity recorded from the reference box (optionally smoothed in time by averaging in a sliding window of arbitrary width) is considered as a raster acquisition to estimate planar displacement using the NoRMCorre algorithm [44]. The estimated displacements are then applied back to the full trajectory. e Schematic of the small and local artefacts correction based on SNR. For each ROI, pixels labelled as belonging to the ROI, its outer ring or its surround are considered. In a sliding window of 10 s the SNR of each pixel is computed, pixels are sorted for decreasing SNR, and only a fixed number of high-SNR pixels is used to extract the fluorescence activity. Pixels are selected line-by-line considering the SNR computed in the subsequent 10 s window