Segmentation of neurons in microscopy images

Set up, train and apply a neural network for segmentation of neurons in microscopy images.

Quick start:

0. Set network and training parameters in Params.m.

1. Prepare input and output images:

Im_In = {Im_In_1,Im_In_2,Im_In_3};

Im_Out = {Im_Out_1,Im_Out_2,Im_Out_3};

2. Generate training set:

Generate_Dataset(Im_In,Im_Out);

3. Train:

net = Train;

4. Apply the trained network to an image:

[Im_Out,Im_Label] = Segment_Neuron(net,Im_In_4);

imshow(Im_Label);

* Example raw and annotated neuron images can be found in this paper [1].

* An example pre-trained network is included.

* Please cite this paper [1].

Advanced options:

- Control sample size (see "Input_Size" in Params.m).

- Control class weights during training (see "Class_Weights" in Params.m).

- Control the minimum number of neuron pixels in training samples (see "Functions" block in Params.m).

- You can generate the training set locally and train on another machine (see "Paths" block in Params.m).

1. Yuval, O., Iosilevskii, Y., Meledin, A., Podbilewicz, B., & Shemesh, T. (2021). Neuron tracing and quantitative analyses of dendritic architecture reveal symmetrical three-way-junctions and phenotypes of git-1 in C. elegans. PLoS computational biology, 17(7), e1009185.