This is a project of the Google Developer Student Club of FPT University Da Nang, built by members, and entered the TOP 10 finalists at the FPT Edu Hackathon 2021 in Hanoi, Vietnam.
Project name: Deep Learning application to detect patients infected with Covid-19 from chest CT images.
Folder Gdrive details about project (includes code file, video demo, presentation) here
Contact me by email if you want any suggestions (dataset, R&D guildline,..): [email protected]
The COVID-19 pandemic is causing major outbreaks in 150 countries around the world, severely affecting the health and lives of many people around the globe. Over the course of our research, we've found that one of the key steps in fighting COVID-19 is the ability to detect infected patients early enough and get them into intensive care. Detecting this disease from an X-ray is probably one of the fastest ways to diagnose the disease in a patient.
Excerpts from a number of international studies show that Cough is the more common symptom and occurs in 68-83% of people infected with Covid-19 and it is the late stage of this symptom that is one of the most common causes. pneumonia leading to patient death. In addition, the actual diagnosis can be complicated because the pneumonia caused by Covid-19 is barely visible or difficult to predict on radiographs, Inspired by previous works, the team studied the application of a deep learning model (Deep Learning) to detect patients infected with COVID-19 from their chest X-ray images.
Up to the present time, Vietnam has recorded many cases of death diagnosed due to severe pneumonia, progressive respiratory failure due to COVID-19. In addition, the unpredictable change of the new SARS-CoV-2 virus variant will increase the number of cases and the possibility of respiratory transmission. Without the application of advanced medical technology research, it will be difficult for doctors to diagnose and perform clinical examination for continuously hospitalized infections. Therefore, the project wants to help experts, doctors and nurses in Vietnam and around the world to overcome the challenges of the COVID-19 pandemic.
The team hopes that the project's research method characteristic can be tested and contributed to the diagnosis and examination process from initial infections to acute pneumonia, especially as a potential contribution. performance in diagnosing multiple cases at the same time.
We used the Image Instance problem to quantify and measure the volume of the pneumonia infection area caused by the COVID-19 virus based on Deep Learning algorithms, basic and advanced Machine Learning methods. . From there, it can provide useful information about fighting the COVID-19 pandemic.
Model towards: U - NET architecture based on Convolutional Neural Networks (CNN) model.
Tools that the team uses include: TensorFlow, Keras, Pandas, Google Collab, PyCharm, 3D Slicer, ...
The team hopes that the project's research method characteristic can be tested and contributed to the diagnosis and examination process from initial infections to acute pneumonia, especially as a potential contribution. performance in diagnosing multiple cases at the same time.
Briefly describe the effectiveness of the product when implemented in practice:
To date, X-ray and CT imaging has been an affordable, common, detailed screening tool that effectively visualizes and accelerates the assessment of severity of COVID-19 lesions. For this research project, the team aimed to demonstrate the utility of an automated tool to segment and measure pneumonia based on computer simulations on data sets. The tool's strengths include its ability to quantify lesions, help visualize infected areas, and quickly monitor disease progression. Furthermore, the team's research holds great promise in its plan to combine imaging data with clinical manifestations and test results to help better test, detect and diagnose COVID-19.
As the COVID-19 pandemic will continue to spread around the world on an unpredictable trajectory, the team hopes that the research can be suggested for large-scale clinical applications; contribute to building cutting-edge tools for health systems, and collectively face similar challenges, including abnormalities caused by viruses and other diseases.
MIT