Development of Neural Network-based Autofocusing Module of Digital Microscopy System for Cytology

• Published in: 2020 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (EIConRus) pp. 2497 - 2502
• Main Authors: Kolokolnikov, George A., Seryogin, Gennady E., Samorodov, Andrey V.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.01.2020
• Subjects: autofocusing; Computational modeling; convolutional neural network; Convolutional neural networks; cytology; Data models; deep learning; digital microscopy; Microscopy; Task analysis; Training
• ISSN: 2376-6565
• DOI: 10.1109/EIConRus49466.2020.9039301

Autofocus is a key function of any microscopy system. Performance of the entire system depends on the autofocus time consumption. The paper is devoted to development of a neural network-based autofocus module of an automated microscopy system. Deficiencies of classical autofocus algorithms are identified and tasks for module development are set. The following stages of module development are considered: choosing neural network architecture; collecting and processing the database; formation of datasets; training and analysis of neural network models; validation of the best model; embedding the model into the autofocus algorithm. The research of various approaches to architecting neural networks for autofocus purpose is carried out. Selected neural network is tested in terms of accuracy and time efficiency. The obtained results are analyzed, and ways of improvement are considered. The results of the paper allow to increase performance of autofocusing, which can be beneficial for biomedical microsample analysis.