Home

Neuron Segmentation in Calcium Imaging

Problem: Identifying the neurons in the given sample.

Calcium imaging is one of the prominent techniques for identifying the activity of neurons over the period of time and for varying amounts of calcium concentration. This process takes a lot of time for labeling the neurons manually and even requires a considerable amount of human inspection despite tremendous efforts to automate the process. The reason for this problem being complicated is because the neurons tend to flicker over time due to increasing addition of calcium.

A single image cannot be used to identify all the neurons present in the given patch, so a series of images are taken as image sequence over some certain period of time.

Data Description:

There were 19 training samples and 9 testing samples with each folder representing unique samples at different times and varying calcium concentrations. Each folder contains a variable number of images.

As folders contain unique samples the positions and number of neurons changes from one sample to another.

The training data consists of coordinates of the image surrounding the neurons. These coordinates are available in the regions subfolder of each training data folder. The regions subfolder consists of a json file where each neuron is represented with an id and list of coordinates entirely surrounding the image.

The data for training and testing can be found at NeuroFinder website with a brief description of the data.

Approaches:

• Naive segmentation
• Non-negative Matrix Factorization
• Neural Networks