Work Package: WP5
Deliverable: 14.1: A robust segmentation algorithm for two-photon calcium imaging
Deliverable due date: month 30
This document reports on the progress of work on Deliverable 14.1, focusing on its relation to the NETT proposal. P14 and its deliverables were modified from those originally expressed in the NETT 289146 as noted in previous reports. From the Mid-term Report, "P14 is developing theoretical/computer vision approaches for the analysis of large scale optical neural recordings".
The developed algorithm is a segmentation algorithm based on each pixel temporal activity average intensity value of a movie acquired with 2-photon microscopy. It exploits the correlation properties between pixels belonging to the same cell combined with the possibility for a human operator to approximately indicate a putative set of core pixels for each region of interest. Some of the key steps of the algorithm involve the computation of a local correlation map, whose entries correspond to the sum of the temporal correlation coefficients between each pixel and its neighbours, and the derivation of normalized correlation maps, i.e. images of the correlation between a set of core pixels and all the other pixels in the movie. For grey-level segmentation we use a graph cut segmentation algorithm.
The algorithm described above is generalizable to a range of scenarios and different data sets. It was developed autonomously but could not be published independently because we discovered later that an equivalent version has already been used in a paper. However, it is suitable for segmenting the 2-photon calcium imaging data that we will collect during the following stages of the project, thus satisfying Deliverable 14.1. The remaining part of P14, “Information theoretic analysis of signals from two-photon calcium imaging” will be reported in Deliverable 14.2.
Contributors: Stefania Garasto, Simon Schultz (Imperial)