Convolutional Neural Network (CNN) for COVID-19 Lung CT Scans Classification Detection

Abstract

COVID-19 disease has been taking away millions of lives from the earth since 2020. Thousands of people have lost their lives because of the shortage of limited medical resources. To tackle this urgent issue, this paper proposed a CNN classification model based on application to help a patient determine if COVID-19 infects lung CT scan of the patient. The application contains a user interface based on the Flask framework, combined with a trained CNN model, which will proceed with the CT scan uploaded by the patient and inform the patient of the result. Additionally, the application simulates a personal virtual assistant, which can be utilized to answer patients' queries. The personal virtual assistant, also known as a chatbot, is made with Natural Language Processing (NLP), which will transfer the patient's input into tokens and return the most likely response to the patient. Unfortunately, after constructing the CNN model, the prediction was not highly accurate. The experimental result demonstrated that the loss value is generally decreasing throughout the training process, and the accuracy score is generally increasing. However, the overall performance was not smooth, which means the model was unstable. The reason that causes this problem is the shortage of the training dataset.