The inspiration came from the lack of quality on the COVID-19 chest X-ray dataset while I was developing an application that detects infected people using an X-ray image. This application was based on deep neural networks and the data that I used for the training process came from the COVID-19 chest X-ray dataset. The trained model achieved almost 80% accuracy but the dataset was really small so these results can't be trusted. It was clear that the training set wasn't big enough and the quality of the scans in some cases was poor. So, I asked myself, how the scientific society can contribute to the creation of high-quality medical datasets? What platform can we develop to help the medical experts provide those data? By having those quality medical data, accurate, fast, and reliable tests can be made, using AI tools.

What it does

We introduce a web application named Hippocrates(ancient Greek and Father of Medicine) that helps experts to estimate the risk of a patient to be infected by COVID-19, using only a chest medical image(X-ray or MRI). The health expert captures the medical image and uploads it to our platform. Then the risk of infection of COVID-19 is estimated within seconds using an AI tool based on Deep Neural Networks and Computer Vision. This application also collects those medical data to create quality medical datasets that can be very useful for research and other experts. By having those new data, we can constantly evolve our Deep Learning model and have much more accurate results on the estimations for COVID-19.  In cases of emergency like the coronavirus outbreak, medical datasets that can be used by researchers and experts can be made within hours.

NOTE: You can enter the site without credentials, type everything you want or just login as Doctor WHO! (The website is not optimized so it needs some time to load, thanks for you patients)

The doctor is also able to write his final diagnosis in our application via text or just record his voice. Then, in our application, we use a sophisticated tool of Natural Language Processing(NLP) that extracts labels(or keywords) from the diagnosis text, or the voice (speech recognition). Those keywords in the case of COVID-19 can describe the symptoms(fever, cough, shortness in breath, etc.) Finally, the extracted labels are linked with the other medical data of the patient and stored in our cloud databases anonymously.

With all kinds of medical data(images, text), several and different datasets can be made. Those datasets can be divided by region/country, age, gender, stage of illness, and of course by the labels we extracted. This feature can be really useful for research scientists and medical experts to be able to access a dataset with specific parameters instantly.

This design can be generalized in the future. The first step is to collect medical image data(CT or MRI scans, etc.), or other medical data(blood test, pressure test, etc.) of patients with COVID-19, pneumonia, cancer, tumors, etc. from several institutes and hospitals worldwide. After these data are provided we can implement Computer Vision  algorithms for the preprocess and quality check of our medical image data, and Deep Learning algorithms to train new models and deploy them. Then, our application will be able to predict the risk of other diseases. When new medical data are provided the models will be updated using transfer learning. This way our AI tools will be constantly evolving and make more accurate predictions. An API can be also developed so other platforms can interact with ours. This tool can help experts to detect disease with more ease because there are pieces of information in the data that are not easily visible to the naked eye. An AI tool could detect them because it is trained in thousands of samples. This tool, of course, can't replace the doctors or the experts and it should be always supervision of the results.

How I built it

For the frontend of this application, HTML5, bootstrap4, and CSS3 were used. For the backend of the web application we used javascript and the AI tool was developed using Python, Jupyter, Keras, and TensorFlow. The website is hosted locally but we exposed it to the internet by using a secure ngrok tunnel. The database we used at this development stage was also local.

Challenges I ran into

The workload for just 2 developers was massive and we were also forced to spend a lot of time in other more important for the hackathon tasks, like composing the business plan, the texts, the pitch video, and the presentation of this project. This is a high complexity project and we need more manpower to bring it to life. This application is using sensitive medical data and not all people willing to provide them for research purposes, even anonymously. What about GDPR? Legal help in those topics is needed. (Edit:We have been advised by a legal mentor that is perfectly legal to continue the development of this project at this point)

Accomplishments that I'm proud of

Putting together a small team of people of different cultures, timezones, backgrounds, and skills in a very short period. This team consists of some new and old friends. Together we composed a project that can make a huge impact on public health by giving the opportunity the experts and organizations, share their knowledge with the rest of the world instantly. This application is dedicated to helping the medical professionals who risk their lives daily through this pandemic. This small team was able to deliver this hackathon project and a demo of our website in just 48 hours with a lot of work and coffee! We would also like to thank publicly our mentors who guided us on the right path. We learned a lot from them in such a small period of time.

What I learned

Much work can be done by the right group of people when they are motivated to build something that can contribute to the greater good. Also mentor-ship was really helpful and some of them   influenced as a lot to get this project in the right course giving us some very good advises.

What's next for Hippocrates Medical Solutions

The next step is to develop everything we weren't able during the hackathon and even more. A nice feature would be the translation of the application to all the EU(at least) languages. This application can be used worldwide from doctors, radiologists, hospitals, and health institutes and I think we will be able to create medical datasets and share our knowledge in cases of emergency like CONVID-19 much faster. Although, there is a big discussion about medical data quality. For example, if the X-ray, CT, or MRI scans don't have good quality this can leed our model to false accusations and predictions. Also, the legal and the business side of this application needs much more development. The key to the feasibility of this project is to have major partners like governments or big health organizations and convince them about the importance of this application and the impact it can make to the lives of millions of people.  The scale of this application depends on how many users are we going to have and how much cloud storage space we need to store our data.  I believe an application like this has to be developed soon and with the right resources and will to work we can bring this project to life!

Note: You can enter the website without credentials, just type something or log in as Doctor WHO! (the website is not optimized so it needs some time to load for the first time, thanks for your patience)

Try It out



ai, cloud, computer, computing, css3, data, deeplearning, edge, html5, javascript, keras, medical, natural-language-processing, python, tensorflow, vision

Devpost Software Identifier