SIMA: Python software for analysis of dynamic fluorescence imaging data

Abstract

Fluorescence imaging is a powerful method for monitoring dynamic signals in the nervous system. However, analysis of dynamic fluorescence imaging data remains burdensome, in part due to the shortage of available software tools. To address this need, we have developed SIMA, an open source Python package that facilitates common analysis tasks related to fluorescence imaging. Functionality of this package includes correction of motion artifacts occurring during in vivo imaging with laser-scanning microscopy, segmentation of imaged fields into regions of interest (ROIs), and extraction of signals from the segmented ROIs. We have also developed a graphical user interface (GUI) for manual editing of the automatically segmented ROIs and automated registration of ROIs across multiple imaging datasets. This software has been designed with flexibility in mind to allow for future extension with different analysis methods and potential integration with other packages. Software, documentation, and source code for the SIMA package and ROI Buddy GUI are freely available at http://www.losonczylab.org/sima/.

Keywords: Python language; analysis software; calcium imaging; fluorescence imaging; in vivo GECI imaging; motion correction; multi-photon microscopy; segmentation.

Figure 1

Workflow supported by SIMA. (1) An ImagingDataset object is first created either directly from the raw data or from the output of the motion correction algorithm. (2) ROIs are generated by automatic segmentation. (3) The ROI Buddy GUI can be used to edit the automatically generated ROIs and to automatically register ROIs across multiple datasets. (4) Dynamic fluorescence signals are extracted from the imaging data and ROIs.

Figure 2

Line-by-line correction of within-frame motion artifacts. (A) Schematic diagram showing a single imaging frame before (left) and after (right) line-by-line motion correction. A separate displacement is calculated for each sequentially acquired line from the laser scanning process. As a result, some pixel locations may be accounted for multiple times (darker blue), while others may not be imaged in a given frame (white gap). (B) Overlay of different regions imaged by different frames due to motion. The light gray region indicates the maximum frame-size that can be selected for the motion correction output, such that all pixels locations that were ever imaged are within the frame. The dark gray region indicates the default and minimum frame-size that can be selected for the motion correction output, such that all pixels locations within the frame are within the field of view at all times.

Figure 3

The ROI Buddy graphical user interface. (A) Image viewing panel with ROI editing tools. During typical use this panel is expanded to occupy the majority of the screen. (B) Panel for toggling between “Edit” and “Align” modes, loading imaging datasets, and registering ROIs across datasets. (C) Panel for selecting, creating, saving, and deleting sets of ROIs associated with the active imaging dataset. In “Align” mode, ROIs from all loaded datasets can be viewed simultaneously. (D) List of ROIs in the currently selected set, and tools for tagging, merging, unmerging, and re-coloring ROIs. (E) Contrast adjustment for the underlying base image. (F) Panel for selection of the underlying base image.

Figure 4

Registration of ROIs across imaging sessions acquired on two different days. (A) ROIs (red) and time-averaged image for the first imaging session. (B) ROIs (green) and time-averaged image for the second imaging session, with ROIs for the first imaging session (red) shown for comparison. (C) Same as (B) but with an affine transformation applied to align the time-averaged image and ROIs from day 2 to those of day 1. (D) Same as (C) but with the ROIs colored by their automatically determined shared identities across both imaging sessions.

Figure 5

Segmentation steps or identifying pyramidal cell nuclei with the ’ca1pc’ variant of the normalized cuts segmentation approach. (A) The time-averaged image of the time-series to be segmented. (B) Application of CLAHE and unsharp mask image processing to (A). (C) Disjoint regions identified by iterative partitioning with the normalized cuts algorithm. (D) Local Otsu thresholding of each region in (C). (E) Cleanup of the Otsu thresholded regions in (D) with opening and closing binary morphology operations. (F) Resulting ROIs after rejection of regions in (E) that failed to satisfy minimum size and circularity requirements.