ABLE: An Activity-Based Level Set Segmentation Algorithm for Two-Photon Calcium Imaging Data

Abstract

We present an algorithm for detecting the location of cells from two-photon calcium imaging data. In our framework, multiple coupled active contours evolve, guided by a model-based cost function, to identify cell boundaries. An active contour seeks to partition a local region into two subregions, a cell interior and exterior, in which all pixels have maximally “similar” time courses. This simple, local model allows contours to be evolved predominantly independently. When contours are sufficiently close, their evolution is coupled, in a manner that permits overlap. We illustrate the ability of the proposed method to demix overlapping cells on real data. The proposed framework is flexible, incorporating no prior information regarding a cell’s morphology or stereotypical temporal activity, which enables the detection of cells with diverse properties. We demonstrate algorithm performance on a challenging mouse in vitro dataset, containing synchronously spiking cells, and a manually labelled mouse in vivo dataset, on which ABLE (the proposed method) achieves a 67.5% success rate.

Automated cellular structure extraction in biological images with applications to calcium imaging data

“…Pnevmatikakis et al [14] presented a state of the art method based on constrained matrix factorization, which explicitly models the calcium indicator dynamics. More recent work has deviated from sparsity-based solutions towards general model-free and data-driven approaches [8,16], and our framework detailed in this paper falls within this line of work. Our extraction does not rely on modeling the fluorescent calcium activity but takes a general approach to solving clustering of high-dimensional data in the presence of noisy clutter.…”

“…Embedding norm. An image summarizing the activity in a calcium imaging video is useful for both manually identifying ROIs, ground truth labeling, overlaying identified ROIs to determine what structure they capture and initializing ROI extraction methods as in [8]. Commonly used visualizations are the temporal mean image, temporal maximum image or a temporal correlation image [9] computed by…”

“…Another possibility to construct l is to provide a user-interface with which to mark suspicious regions, for example, by providing the user with the maximum embedding norm, however this is outside the scope of this paper. Either ROI candidate initialization proposed in [8,14] can also be applied to construct l. Since we iterate only over points with high embedding norm, we essentially ignore the clutter.…”

“…A second challenge is the images themselves, which suffer both from varying dynamic range, and low signal-to-noise ratio due in part to the activation of calcium in the neuropil creating a noisy heterogeneous background [7]. In addition the data suffers from measurement noise and movement artifacts [8]. Analysis of calcium imaging data has typically relied on the assumption that ROIs are spatially sparse (occupying a small region of pixels in the image plane), with a sparse temporal signal.…”

Low Computational-cost Cell Detection Method for Calcium Imaging Data

“…Marius et al (2016) proposed a NMF based method called Suite2P, which contains a model of background fluorescence changes due to neuropil activities. Different from these NMF methods, Reynolds et al (2017) proposed ABEL, a method based on an active contour model. It has been reported that ABEL outperforms CNMF and is comparable in performance to Suite2P.…”