Title: Automated analysis of blood vessel permeability using deep learning
Authors: Yoojin Chung; Hyunwoo Kim
Addresses: Division of Computer and Electronic Systems Engineering, Hankuk University of Foreign Studies, Yongin 17035, South Korea ' Division of Computer and Electronic Systems Engineering, Hankuk University of Foreign Studies, Yongin 17035, South Korea
Abstract: This paper presented a deep learning-based approach using CNN that automatically detects changes of permeability in the blood vessel network after exposing the vessels to chemicals. Because chemicals can affect the permeability of the vessels, it is important to characterise and quantify the changes to blood vessels accurately from time-lapse microscope images. We increased the number of data by applying diverse augmentation methods to 2,800 images. In this paper, we also devised a method of automatically labelling images as to whether there is any change in the blood vessel to obtain learning data for CNN by pre-treating images using diverse image processing methods and separately establishing thresholds for different images. To verify the performance of the developed CNN, we used 4-fold cross validation and blood vessel changes were automatically detected with an accuracy of 92.07% in the experiment.
Keywords: blood vessel network; automated recognition; microfluidic culture platform; convolution neural network; CNN; blood vessel permeability.
International Journal of Internet Technology and Secured Transactions, 2021 Vol.11 No.4, pp.341 - 351
Received: 08 Apr 2020
Accepted: 13 May 2020
Published online: 01 Aug 2021 *