SDG 3: Good Health and Well-being
SDG 10: Reduced Inequality
2. Project Details
Company or Institution
National Cheng Kung University
MedCheX: An e-Alert system for automatically detecting pneumonia from chest X-rays
General description of the AI solution
As we continue to face the rapid increase in confirmed Coronavirus cases around the world, our team created an AI pneumonia detection platform for suspected COVID-19 patients. The purpose is only takes a single second to assist front-line doctors in recognizing patients who are infected.
The precise and rapid diagnosis of COVID-19 in the early stages of the disease is of utmost importance. Although doctors may use RT-PCR tests to help diagnose the disease, it is a complex, time-consuming, and costly process. COVID-19 pneumonia aggravated in rapid progression stage presenting bilateral multi-lobe light consolidations with air-bronchogram inside that could be detected early on chest X -rays. Chest X-ray is still one of the most effective, fast, and cheap clinical screening tools.
While suspected COVID-19 patients in the test station of domestic hospital have been waiting in the queue for too long, the system is built to automatically detect high-risk patients with pneumonia and/or COVID-19 that will then send the visual symptoms information to doctors. With that information, doctors are then able to make decisions and provide a treatment plan for the diagnosis. Any X-ray images taken for patients could automatically be uploaded to our system. The AI system will then scan the images to determine whether the patients are possibly infected.
The automatic pneumonia detection system, named "MedCheX," developed by our team has become one of the winners of the online competition organized by the World Health Organization (WHO) and dozens of high-tech companies globally to seek solutions for tackling challenges related to the COVID-19 coronavirus pandemic last year. The team was announced as one of the 25 highlighted projects among 1,560 proposals from 175 countries in the COVID-19 Global Online Hackathon.
National Cheng Kung University
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
Among various types of prediction models, convolutional neural network (CNN)-based models have significant successes in a variety of computer vision tasks such as medical image analysis and object detection. The main advantage of using CNN-based models is the low number of parameters and hyper-parameters of the model that helps to speed up the models training, increase the accuracy, and prevent the over-fitting.
In this AI platform, we designed a real-time CNN-based model, MedCheX, to assist frontline physicians, radiologists, and nurses for quickly recognizing pulmonary infiltrates on chest X -rays. The model distinguishes COVID-19, infected lung, patterns from normal medical images. Besides, it precisely segments the lesion areas even in low resolution images. The MedCheX model can identify the lesion area and automatic label them into at least three groups: consolidation, ground-glass opacity, and retrocardiac. These lesion areas are demonstrated to the physicians to let them make crucial decisions based on advice guideline for the diagnosis of COVID-19 infected pneumonia.
The MedCheX model is based on a well-known CNN-based model, U-Net++, that uses multiple down-sampling and up-sampling techniques to capture informative features. It utilizes DenseNet and skip connection techniques for encoding the internal features. Finally, we used Adam optimization technique to minimize the prediction error during the training phase. We performed a two-phase training scenario using more than 4000 pneumonia X-ray images from National Cheng Kung University Hospital to train the AI classifier. We achieved 0.868, 0.920, and 0.893 for precision, recall, and F1-score, respectively. We compared our proposed model with well-known previously published models including original U-Net++ and U-Net++ with SE blocks. Interestingly, our MedCheX model outperforms the other models.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
This proposal primarily focused on “SDG 3: Good Health and Well-being” to ensure healthy lives and well-being for all at all ages. In addition, we also consider “SDG 10: Reduced Inequalities”. Every country develops at a different pace, so free of charge policy allows countries with different conditions to easily access this system. Amid this pandemic, every individual and country should be treated equally, no one is left behind because of any reason.
We are honored to be one of the 25 winning teams selected from 1,560 proposals in the COVID-19 Global Hackathon, and it is also our desire to freely offer up our solution globally. We understand that not every country may be as well off as developed country is currently and we want to do anything possible to help mitigate the health crisis that is facing us globally. To that effect, we feel that making our project (MedCheX) available to any medical professional that wishes to use it will not only foster increased collaboration on such projects in the future, but also help to ameliorate our current situation. This may be especially beneficial for countries that do not have enough medical resources or radiologists. With this system, even in remote areas where there is a shortage of skilled doctors, the AI system can still quickly detect and prevent the spread of the epidemic.
According to the Google website public analysis tool GA as of the end of May, there are thousands of users from 61 countries all over the world who have been using the MedChex platform. That’s quite impressive number.
We are so proud to be one of the teams that create positive impacts on United Nations. We hope that more people will use it so we can help. Together we can make the world a better place.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
Currently this platform can be used to analyze chest X-ray (CXR) images in test stations and/or outpatient clinics for any hospitals easily. A physician can upload CXR images to the online AI platform through the Internet. Then, the CXR images are being interpreted by MedCheX immediately to highlight the suspected lesions for the physician.
Any medical professional around the world can register in the platform by inserting user name and password. Then, after logging in to the system, the user can upload a single or multiple CXR images. After model computation, the results including the original and the suspected infection heatmap images appear in the screen.
According to the Google Analytics tool as of the end of May, medical staff or radiologists from 61 countries use this web-based MedChex AI detection system. Through the Internet, these medical staff can directly upload the X-ray images to our platform. Our system then automatically assists in interpretation and immediately returns results from the AI model, predicting relevant information for the users. Those users come from large and small countries, whether those with access to advanced healthcare or those with more limited resources. Included in Asia, for example: India, Japan, and South Korea, as well as and Turkey and Vietnam; in Europe: Italy, Luxembourg, Netherlands, and Slovenia; in the Americas: several hundred users in the United States alone, as well as Brazil, Chile, and Guatemala in South America; in Africa: Kenya, Nigeria, Tunisia, etc.
Imagine there's best experienced radiologist from New York can help India people to fight COVID-19 ? We need more doctors to use it, and we can also help more medical staff. We further add a section to make the possibility of sending user feedback to the AI platform to help the AI model improving its accuracy in the future.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
We came up with the decision to provide medical staff around the world with the MedCheX system free of charge web service. In light of strategy, regardless of developed or disadvantaged countries, doctors can easily screen for COVID-19 without experience in pneumonia diagnosis. This makes the development of our system has to minimize risk and ensure platform continuity by follow the requirements of the Information Security Management System (ISMS).
During the development of the MedCheX platform, we make sure that we follow our own Institutional Review Board (IRB) to secure the safety of image data of participants and prevents violation of human rights. Our basic ethical considerations are the principles that must be followed in conducting our preliminary research. Our university has our own Code of Ethical Practice, and it regulates the issues of privacy and data protection specifically a set of policies and procedures for systematically managing the sensitive image data. It is critically important for us to thoroughly adhere to this code in every aspect of our work and declare our adherence in ethical considerations part of our MedCheX platform, which included medical image consent, data handling, data collection, data processing, and training of the AI models.
This ethical aspect does not mandate the development of our AI system, but includes suggestions for documentation, internal audits, and continual improvement of the platform.