The acute Covid-19 pandemic is becoming more and more threatening every day. All of us are having unimaginable tough times. Every day someone is losing loved ones (family or friends), others losing their jobs, and all of us are losing freedom of movement. We live at a time of many opportunities and we must not allow this to happen again. The horrible aspect of the Covid-19 pandemic is nobody knows when it will end. Will it return again? We should not stand and wait but we must act NOW.
There is a huge lack of test kits and medical resources to control the fast COVID-19 spread. Authorities are not able to identify asymptomatic or early symptoms of infected people before any call to health services. Health services are overloaded by a large number of calls. The new virus comes with many unknowns such as symptoms, treatments, tests, etc. Furthermore, there is no open health data set to help medical staff, researchers and other institutions to accurately make data-driven based decisions.
To addresses and solve the aforementioned concerns we build an AI-based platform which can be used:
- as a pre-diagnosis tool to assess people's risk level of COVID-19 (low/medium/high),
- to accurately identify high-risk Covid-19 areas,
- track symptoms and health conditions of people,
- by experts in various medical and other fields to explore our open health data to give their support in dealing with Covid-19 crises.
What it does
The Darint AI platform combines three functionalities into one place:
- SymCough checker, a pre-diagnosis tool to evaluate a risk level (low, medium, high) of being infected with Covid-19,
- Symptom tracker to track the patients' symptoms anonymously,
- Open Data Dashboard, an analytical dashboard with all anonymized collected data. The patient in order to evaluate the risk of being infected with Covid-19 will be guided through our Darint platform to use SymCough checker to answer some symptoms related questions and record cough sounds (using the smartphone or PC). The SymCough checker will then process the data and evaluate a risk level (low, medium, high) of being infected.
There is various scientific evidence that indicates that cough can be used to detect respiratory diseases including Covid-19. However, according to WHO, around 30% of patients have no cough symptoms. Therefore, using a cough-based detection method alone does not fulfill healthcare standards. On the other hand, online symptom checker services demonstrate increasing accuracy.
- First, we build a cough-based detector using the Convolutional Neural Network achieving 89.72% accuracy.
- Next, we build a symptom checker using the Bayesian networks and the knowledge graph achieving 93.6% accuracy.
Then, we build a SymCough checker which combines cough-based detector and symptom checker, achieving an accuracy of 96.41%.
At the time of writing, as per our knowledge, our solution has the highest accuracy of 96.41% compared to others in this domain. On the other hand, our Symptom tracker continuously improves the accuracy of the SymCough checker, while all anonymized collected data are open in the Open Data Dashboard and can be used by other researchers and institutions.
Simple to Use and Easy to Scale to millions of people: everyone, in every Country in any Language can use it.
Our solution is applicable to other diseases and any future pandemic situation.
- increase user satisfaction with the use of mobile technologies to track their symptoms and to evaluate the risk of being infected without affecting their privacy,
- be informed about infected areas with Covid-19.
- protect themselves from being infected from people evaluated at high risk by our platform,
- analyze the historical patients’ symptoms.
The healthcare local communities, healthcare institutions, and national authorities:
- manage the situation better,
- serve more people, while accurately identifying high-risk Covid-19 infected areas,
- reduce calls and stress,
- eliminate false-positive Covid-19 test kits,
- evaluate any post Covid-19 consequences on patients.
Our platform will be offered at no cost for Covid-19 pandemic. However, it can be easily monetized and used by hospitals to pre-diagnosis their patients and track and analyze their symptoms.
How I built it
- Backend: Php, Python, Flask, Flask API
- CMS: Wordpress
- Database: MySql
- REST APIs
Challenges I ran into
- Lack of health labeled data.
- Clinical trials.
Accomplishments that I'm proud of
In a very short time, we build a platform that can help the world to deal better with unexpected situations like a Covid-19 pandemic.
What I learned
We learned that in crisis time it is important to share all of the data that is related to the crisis so that all can work together to minimize the consequences of the crisis.
What's next for Darint
Extend our solution to detect asymptomatic Covid-19 cases using pulse measurement and heart rate variability.
Endorsement and validation of our solution by healthcare institutions and national authorities, in order to make our solution recognized and trusted by people.
Partnership with Healthcare Institutions.
Integration of our platform with health tracking devices.
Time to full deployment
Although our platform is live and working, to be ready to be used as a reliable platform by people it depends on the availability and usability of symptoms and cough sounds datasets and the possibility to cooperate with healthcare providers to train our AI platform. Additionally, our platform depends on medical evidence currently being investigated on the validation of cough sounds and symptoms as a reliable indicator for the detection of Covid-19.
- Dardan - Machine Learning Consultant
- Valon - Software Web Engineer
- 2 external advisors with a medical background