کنفرانس مهندسی برق ایران

صفحه اصلی / سی امین کنفرانس بین المللی مهندسی برق

An Investigation on Transfer Learning for Classification of COVID-19 Chest X-Ray Images with Pre-trained Convolutional-based Architecture

نویسندگان :

Mobina Abdoli Nemati¹ ََAmirreza Baba Ahmadi²

1- دانشگاه علم و صنعت ایران 2- دانشگاه تهران

کلمات کلیدی :

Deep Learning،Transfer Learning،COVID-19 Detection،Medical Computer Vision،Medical Image Analysis

چکیده :

Medical image analysis techniques has been found useful for rapid identification of COVID-19. Since it's difficult to find important visual features for a large number of patients, machine learning methods have been found promising to diagnose whether one is infected or not. Since, there are many approaches of using convolutional-based architectures for medical image analysis tasks, we have used the most famous and potent deep neural networks to provide a concise and reliable source of information on transfer learning-based methods to open the road for more advanced studies. This article presents an investigation on 27 pre-trained convolutional neural networks to recognize the existence of COVID-19 in medical chest X-ray images. In this study, we have used a binarized version of the few public datasets from COVID-19 patients and non-COVID subjects. The evaluated results from 5-fold cross validation show that architectures such as: EfficientNet networks, MobileNet, Inception-ResNetV2, NasNet-Large give a superior performance on the aforementioned binary pattern recognition task.