DIMAI is a project created by a multidisciplinary team from Colombia, Mexico and Iceland, we have seen how families in our countries do not have the resources to take a Covid test, where you have to pay high prices and wait days or even weeks to receive the test result, and that is why we are working togueter with the aim of offering hospitals rapid diagnoses, without previous capacitation, without cutting medical personnel, with high precision and a low cost to all the countries around the world.

What it does

The system use artificial intelligence with a large number of chest x-ray images from hospitals and universities to find patterns printed on them, with the help of this analysis and deep learning, the artificial intelligence can give a fairly assertive diagnosis of pacient pathology witg just analyzing the x-ray. DIMAI analyzes the images and can detect if a patient has COVID-19 or other lung diseases, such as:

• Atelectasis • Pleural effusion • Pneumonia

And of course, healthy patients.

How we built it

For greater efficiency when used by hospitals, we decided to create a web platform in django, where users can use the system from any device connected to the internet, the system is linked to a neural network previously trained in python which receives the image of the patient and makes the analysis with the images of the database. After all this process, the medical professional receives the diagnosis from the patient.

Dimai was created to be completely user-friendly and easy to use, no training is required to use the system and also we add a chat bot created in dialgoflow to guide and answer the new user questions.

And the system architecture we use the MVT model for greater reliability for maintenance, updating and flexibility to changes that allow us to continuously improve the system model.

Challenges we ran into

The first problem we encountered was the collection of images of raffiagraphies to enter the neural network. In the last week we have accumulated a database of more than 100,000 radiographs and with 15 different pathologies, including those already specified.

The second most important that we have had is the hardware necessary to trian the neural network where we need a high GPU processing power for the immense amount of images that we have, this problem is also reflected in the web server where the server needs the capacity to run a previously trained neural network to achieve the expected results.

The third is the effectiveness of the algorithm, developing an algorithm capable of having an effectiveness of over 70% has been a complete challenge, the database is very large and there are biases in the data that cause problems when accurately predicting a diagnosis, we have investigated several algorithms of neural networks and continue to find improvements to make the system more precise and actually we have 50%.

The fourth problem is the investigation of the business model of the project, currently and as an objective we have contemplated the countries of Mexico, Colombia, Iceland and Europe as the initial market, where our biggest consumer would be the private hospitals that could contract the service more easily. In addition we decided for an on-demand model to reach the largest number of hospitals possible.

Accomplishments that we're proud of

We are proud of the work that we have done during this hackathon, we have developed a large part of the Web platform, we have made significant advances in the improvement of the artificial intelligence algorithm, we created a chatbot for new users of the system and we have improved our business model to expand our market.

And we are also proud of what this project and what it can achieve, we are not only fulfilling the need to carry out massive tests, but it will also help many people during and after this crisis, help the most needy countries, standardize a new way to diagnose lung diseases and help reduce the number of infections of medical personnel all countries in the world.

What we learned

We have learned many things developing this project, we have learned from the capabilities of neruonal networks and their interaction with massive amounts of data, we have improved our development skills and we have improved our analysis capabilities.

And we have also been able to learn to work as a team for a common goal, trust others and collaborate with people from other countries, despite the language and the schedule.


We have high expectations of our project, and in the short term we will continue with the development of the algorithm to increase its effectiveness, we will finish a functional prototype of the web platform to test with our partners in Colombia as soon as possible, we will start visiting hospitals, universities and more centers of businesses that want to help our project, and finally we will continue to find colleagues who want to continue helping with our project.

Are you interested in our project or would you be interested in collaborating with us? Please contact us, we will be very happy to contact you.

Built With

Share this project: