USING DEEP LEARNING MODELS TO DETECT PARASITES EARLY

Abstract

In today’s engineering applications, applications that think and behave like humans are emphasized. The naming used for the human phenomenon to take place in engineering applications is known as machine learning. Machine learning is used in many areas such as increasing speed and quality, security applications, classification, medical diagnosis and diagnostic applications, and predictive approaches for the future. Convolutional neural networks (CNN) are known as multilayer neural networks. Important studies have been carried out with this neural network system and successful results have been obtained. With convolutional neural networks, significant works have been carried out in many areas such as signal processing, video analysis, image analysis and detection, classification, medical image processing. While using this neural network, some steps are performed. These are defined as pre-processing, feature extraction and classification-detection. At each stage, special approaches are exhibited and studies are carried out to increase accuracy. Intestinal parasitic infections have been recognized as the most important cause of diseases by the World Health Organization (WHO). Early diagnosis of these diseases is very important after the rapid spread of intestinal parasites. Today, machine learning approaches are used in the early detection of diseases because they provide faster results and less cost. In this study, VGG16, ResNet50, and Inception-V3 network architectures were selected for the classification of microscopic images of intestinal parasite infections, capillaria philippinensis (roundworm), and enterobius vermiculari (pinworm). Data enlargement, normalization, and resizing of image data, which are pre-processing techniques, were applied to the data set. After data pre-processing, classification was made with VGG16, ResNet50, and Inception-V3 networks, respectively. As a result of the experimental studies, it was seen that the VGG16 network achieved the highest success rate with 98.07% Train Accuracy and Test Accuracy.