Automatic segmentation of odor maps in the mouse olfactory bulb using regularized non-negative matrix factorization

Abstract

Segmentation of functional parts in image series of functional activity is a common problem in neuroscience. Here we apply regularized non-negative matrix factorization (rNMF) to extract glomeruli in intrinsic optical signal (IOS) images of the olfactory bulb. Regularization allows us to incorporate prior knowledge about the spatio-temporal characteristics of glomerular signals. We demonstrate how to identify suitable regularization parameters on a surrogate dataset. With appropriate regularization segmentation by rNMF is more resilient to noise and requires fewer observations than conventional spatial independent component analysis (sICA). We validate our approach in experimental data using anatomical outlines of glomeruli obtained by 2-photon imaging of resting synapto-pHluorin fluorescence. Taken together, we show that rNMF provides a straightforward method for problem tailored source separation that enables reliable automatic segmentation of functional neural images, with particular benefit in situations with low signal-to-noise ratio as in IOS imaging.

Automated Functional Analysis of Astrocytes from Chronic Time-Lapse Calcium Imaging Data

“…Some methods rely on priors about morphological properties such as size and shape patterns (Valmianski et al, 2010), and thus are very likely to fail for the astrocyte problem due to the irregular and heterogeneous morphologies of FIUs. The most well-accepted set of sophisticated tools is grounded on matrix decomposition techniques such as independent component analysis (Reidl et al, 2007; Mukamel et al, 2009), sparse dictionary learning (Diego et al, 2013), multilevel sparse matrix factorization (Andilla and Hamprecht, 2013), non-negative matrix factorization (Maruyama et al, 2014; Soelter et al, 2014), and constrained non-negative matrix factorization (Pnevmatikakis et al, 2016). However, just like CellSort, all these neuron-targeted algorithms make either an explicit or implicit assumption that pixels in an FIU are synchronized under the given imaging resolution.…”

“…To date, a handful of algorithms have been developed for analyzing neuronal Ca 2+ imaging data (Reidl et al, 2007; Mukamel et al, 2009; Smith and Häusser, 2010; Valmianski et al, 2010; Andilla and Hamprecht, 2013; Diego et al, 2013; Pachitariu et al, 2013; Kaifosh et al, 2014; Maruyama et al, 2014; Soelter et al, 2014; Pnevmatikakis et al, 2016). However, none of them can be applied to astrocytic Ca 2+ data due to the specific challenges mentioned above.…”

Computational exploration of molecular receptive fields in the olfactory bulb reveals a glomerulus-centric chemical map

“…We segmented the functional image series into individual glomeruli and their activation course by means of regularized non-negative matrix factorization (rNMF) 36 . Such a factorization disaggregates the measurement matrix into components with a spatial signal distribution D and a common activation course D of the participating pixel.…”

“…We decomposed IOS image series into 150 components with fixed smoothness regularization parameter GH = 2. The sparseness regularization parameter G; was adjusted such that spatial component correlation was just below 0.5 (see Soelter et al, 2014, for a justification of this value). In image analysis performed to extract MOR18-2 responses, we dismissed the usual non-negativity constraint on the components' activation courses to capture possible odour induced inhibitions.…”

“…Signals from individual glomeruli were then extracted using regularised Non-negative Matrix Factorisation (rNMF) ( Fig. 1d-g, see methods and 36 ). The obtained glomerulus-like image components closely matched the original glomerulus outlines obtained from the GFP signal ( Fig.…”