Fast-scan cyclic voltammograms (FSCV) raw data were obtained from the Laboratory of Central Nervous System of the Federal University of Parana (UFPR) at Curitiba, Brazil, and from D. Robinson’s Laboratory of the University of North Carolina (UNC) at Chapel Hill, United States of America. The images were generated from 30 different experimental records with a total of 1005 electrically evoked dopamine release. Each record has dopamine release evoked with different magnitude of electrical stimulation, resulting in different pattern of form and intensity. All experiments were performed in accordance with the NIH Guide for the Care and Use of Laboratory Animals with procedures approved by the Institutional Animal Care and Use Committee of the University of North Carolina, and the Institutional Ethics Committee for Animal Experimentation of the Federal University of Parana (Protocol 638).
This new dataset version is composed of images generated from all these experimental records, which include those in the version 1 – 2018. Unlike the first version, the new one has not only phasic dopamine release images, but also entire images with no DA release. In total there are 2010 images, 1005 of each of these classes, with resolution of 875×656 pixels representing a 20 seconds recording. During the generation of these FSCV images, a background subtraction is commonly used before applying a fake color palette. Normally for each image, one column or more columns are selected to subtract the values from the others. In the case of the first version of the dataset, this process was done manually during its generation. In this new one, the process is done automatically, choosing 3 different background positions: the Background A was selected from a column at the beginning of each image (0.5 seconds), the Background B from the middle of each image (10 seconds), and the Background C from the end of each image (19.5 seconds).
These images with different background end up generating different results. Thus it is possible to explore different approaches of training and testing, since for each DA release 3 images were generated. Each image containing DA was manually labeled with the approximated information of each release interval and peak. All images were divided into 3 folds with same amount of samples from each of the two classes: (1) phasic DA release images and (2) non-release images. The full details are in our papers [ https://doi.org/10.1016/j.compbiomed.2019.103466 ] and [ http://dx.doi.org/10.1109/IWSSIP.2018.8439339 ].
HOW TO OBTAIN ACCESS TO THE IMAGES
The database may be used for non-commercial research provided you acknowledge the source of the image by citing the following paper in publications about your research:
- Matsushita, G., Sugi, A. H, Costa, Y., Gomez, A., Cunha, C., Oliveira, L. S., Automatic dopamine release identification using convolutional neural network, Computers in Biology and Medicine, 114, 2019 [pdf]
Please contact Gustavo Matsushita (gustavomatsushita AT gmail.com) or Luiz Oliveira (luiz.oliveira AT ufpr.br) with questions or comments.
Version 1 – 2018
The dataset of cyclic voltammogram (CV) images was obtained from the Laboratory of the Central Nervous System of the Federal University of Parana (UFPR). The data were extracted from rats with a carbon fiber electrode using a machine for electrochemical records of fast-scan cyclic voltammetry, and stored in a numerical matrix, which was processed and transformed in images with resolution of 1200×900 pixels.
Each information collected in the experiments is represented by the applied potential on the y-axis, the x-axis is the cycle (time), and the color is current. The colors are based on a false color palette agreed by researchers and used by FSCV analysis softwares.
The dataset consists of 9 different experimental recordings with a total of 285 phasic DA releases. Each record has releases with different concentration, that results in images with different intensities and sizes of these events. Image patches with size of 100×100, 120×120 and 150×150 pixels were created using release moments labeled informations and randomly divided into 3 folds. The full details are in our papers [ https://doi.org/10.1016/j.compbiomed.2019.103466 ] and [ http://dx.doi.org/10.1109/IWSSIP.2018.8439339 ].
CONTACT
Gustavo Matsushita (gustavomatsushita AT gmail.com) or Luiz Oliveira (luiz.oliveira AT ufpr.br).