ABSTRACT
Dopamine has a major behavioral impact related to drug dependence, learning and memory functions, as well as pathologies such as schizophrenia and Parkinson's disease. Phasic release of dopamine can be measured in vivo with fast-scan cyclic voltammetry. However, even for a specialist, manual analysis of experiment results is a repetitive and time consuming task. This work aims to improve the automatic dopamine identification from fast-scan cyclic voltammetry data using convolutional neural networks (CNN). The best performance obtained in the experiments achieved an accuracy of 98.31% using a combined CNN approach. The end-to-end object detection system using YOLOv3 achieved an accuracy of 97.66%. Also, a new public dopamine release dataset was presented, and it is available at https://web.inf.ufpr.br/vri/databases/phasicdopaminerelease/.