Comparison of Transfer Learning vs. Hyperparameter Tuning to Improve Neural Networks Precision in the Early Detection of Pneumonia in Chest X-Rays

Abstract

The WHO (World Health Organization) has updated a publication where they discuss the topic of Covid-19 vaccination for children, and in this document, they mention the vulnerability that this illness has caused among children under the age of 5, exposing them to a higher risk of other diseases such as pneumonia. For this reason, this research is focused on the early detection of pneumonia using children’s chest X-rays and the implementation of artificial intelligence. CNN (Convolutional Neural Network) is the best tool to use as an image processor for the chest X-rays, hence a variety of deep learning techniques were used such as VGG16, VGG16-W VGG19, VGG19-W HT, ResNet50, ResNet50-W, MobileNet, and MobileNet-W. To enhance the accuracy of these deep learning techniques, transfer learning, and hyperparameters were applied to the training process. As a result of this research, we’ve obtained an accuracy of 0.9684 and a loss of 0.0793, and hope that with this research we can help the medical areas in the early detection of pneumonia and save doctors time.

Ähnliche Arbeiten