DeepVID: A Self-supervised Deep Learning Framework for Two-photon Voltage Imaging Denoising

This repository is the implementation of the self-supervised denoising model for voltage imaging data (DeepVID), as presented in the publication:

Platisa, J., Ye, X., Ahrens, A. M., Liu, C., Chen, I. A., Davison, I. G., ... & Chen, J. L. (2021). High-Speed Low-Light In Vivo Two-Photon Voltage Imaging of Large Neuronal Populations. bioRxiv.

https://www.biorxiv.org/content/10.1101/2021.12.07.471668v1

Figure 1. Training strategy and network structure of DeepVID.

Figure 2. Example of single frame image denoising with DeepVID.

System Requirements

requirement.txt lists dependencies for the repository. The code should be compatible with packages in later versions.

• Python 3.8

• Tensorflow 2.3.0

• NumPy, SciPy, Matplotlib, scikit-image, scikit-learn, h5py, tqdm

The repository was tested on CentOS 7.9.2009 with a Nvidia P100 GPU.

Installation Guide

This repository requires Python 3.8. If you have Conda, create a virtual environment tf2-gpu:

$ conda create -n tf2-gpu python=3.8

Activate the virtual environment tf2-gpu and install dependencies:

$ conda activate tf2-gpu
$ pip install -r requirement.txt

The installation should be less than 5 mins.

Demo

Training new models

• The script ./scripts/dipn2v_multi_train.py contains the code to train the denoising model.

• The training data should be stored in the path ./datasets/training.

• To begin training, run dipn2v_multi_train.py:

$ python ./scripts/dipn2v_multi_train.py

Denoising with a trained model

• The script ./scripts/dipn2v_inference.py contains the code to denoise with a trained model.

• The image to be denoised should be stored in the path ./datasets/.

• To begin denoising, run dipn2v_inference.py:

$ python ./scripts/dipn2v_inference.py