click here to download the abstract project
click here to download the base paper
ABSTRACT
The COVID-19 pandemic has created much damage to society and brought panic all around the world. This deadly SARS-CoV-2 virus is highly contagious and has affected almost all countries. To stop spreading this virus, quick and accurate diagnosis followed by effective isolation and patient treatment is mandatory at the early stage of virus breakouts. The disease is currently confirmed using reverse transcription-polymerase chain reaction (RT-PCR) testing. However, through research and other reports, it is confirmed that the sensitivity of RT-PCR might not be high enough for early detection and treatment of patients who are assumed to have to have COVID-19. In recent times, artificial intelligence using deep learning technology in the medical imaging domain has excellent success. Deep Learning was applied to efficiently detect and accurately differentiate between bacterial and viral pneumonia in chest CT scans.
CT Scan is an imaging approach used to identify the characteristics and effectively serve for early diagnosis and screening of COVID-19. Detection of COVID-19 Using Deep Learning CT Image Classification using convolutional neural networks: Improving Testing with Deep Learning and Computer Vision.
In this project we shall discuss, how testing is done for the covid-19 and how deep learning tools will be useful for medical imaging can help us in improving the testing quality of COVID-19. We aim to research and implement a Deep Learning classifier that contains the information of COVID and non-COVID patients, which can classify patients based on which we declare the results of the patients with utmost accuracy. The sensitivity of predicting is high enough to detect at the early stage of the virus. As it is a Deep Learning model, the processing time to give the testing results is very fast.
INTRODUCTION
In 2019, Chinese health authorities got to know a new unknown origin of pneumonia.The name SARS-CoV-2 (severe acute respiratory syndrome) means the lungs and respiratory tract are highly affected, leading to the cause of novel coronavirus. This virus was originated from bats and spread to people through an unknown origin. Severe corona symptoms are difficulty breathing, chest pain, and several other common and mild symptoms like cold, headache, fever. At the beginning of the coronavirus, an RT-PCR test (Reverse Transcription polymerase chain reaction) was used to predict the virus. This method is inaccurate because it gives false positives, which means people who are not infected by the virus are told that they have been tested positive, and false negatives, which means people who have the virus, are classified as harmful by our algorithm. This way of wrongly classifying is alarmingly shortcoming as it would allow many infected people to go home and spread the virus. RTPCR results lead to an increase of
spreading the virus as testing consumes more time, whereas now there are computer based models to predict the values and give accurate results. For the RT-PCR test, all the collected samples are delivered to testing in less than two days; else, there is a considerable possibility of false results. The accuracy decreases with an increase in time. Hence this method is not reliable. As a non-invasive imaging approach, CT-Scan can depict specific characteristics manifestations in the lung associated with COVID-19. Therefore, CT-Scan could serve as the most accurate way for early diagnosis of COVID-19. We used the VGG-19 with batch normalization as our inspiration model. We shall be constructing our own neural network model and shall be training the model with our training data. ur aim is to improve accuracy and metrics such as sensitivity, specificity and Area under the curve of the preexisting model. We will be using KERAS and TensorFlow created by the Google Brain team, TensorFlow is an open-source library for numerical computation and large-scale machine learning. TensorFlow bundles together a slew of machine learning and deep learning (aka neural networking) models and algorithms.
TensorFlow is mainly used for: Classification, Perception, Understanding,Discovering, Prediction and Creation.
The main goal of this paper is to improve the accuracy of true positive and negative
results.
