Investigation of the effectiveness of transfer learning and fine-tuning methods on the VGG16 convolutional neural network in the classification of Covid-19, pneumonia and healthy radiographs on chest X-Ray images
Abstract
Investigation of the effectiveness of transfer learning and fine-tuning methods on the VGG16 convolutional neural network in the classification of Covid-19, pneumonia and healthy radiographs on chest X-Ray images
Incoming article date: 24.04.2023In this paper, we investigated the effectiveness of an intelligent system for classifying radiographic images into three classes using transfer learning and fine-tuning methods. The convolutional neural network VGG16 structure was used to evaluate the effectiveness of predicting the presence of characteristic pathologies by the system. A catalog consisting of 4,228 chest X-ray images was used for training, verification of training and testing this neural network. A catalog was divided into three classes such as pneumonia, Covid-19 and in the absence of lung diseases. To provide modeling processes there were used deep learning libraries such as Keras and TensorFlow. In the course of the work, there were presented the results of the neural network accuracy of the classification of chest X-ray images by full-trained neural network model. We also demonstrated an increase in prediction accuracy by using transfer learning method, as well as with fine-tuning of the neural network structure. Based on the results of the work, the neural network has learned to recognize images with signs of pneumonia, Covid-19 and normal radiograph. In conclusion, the best accuracy index of 98.4 % was achieved by using the fine-tuning method for our convolutional neural network.
Keywords: chest X-ray, Covid-19, pneumonia, intelligent system, machine learning, deep learning, convolutional neural networks, transfer learning, fine-tuning of convolutional neural network