We came up with this idea because of the lockdown caused by the coronavirus pandemic people are confined to your homes and it's a bit risky visiting the hospitals as there are chances of getting infected. We all are facing problems because of coronavirus pandemic where people are in a state where anyone suffering from a disease is thought to be suffering from COVID19. Therefore we wanted to help people reduce this state of panic and thought of building a bot which will answer people’s questions regarding symptoms. That way people will gain knowledge of their disease.
What it does
RoboDoc is a messenger bot where people could chat with it and get their symptoms analysed using messages. Based on symptoms sent by user it analysis and most accurate diseases is diagnosed. At present we have added 21 common diseases we will be expanding it to 87 diseases our main objective was to lessen the panic caused by coronavirus pandemic. We have added analysis of Frontal chest x-ray for covid19 and analysis of mamograph for breast cancer detection.
For COVID19 detection model using X-rays we used kaggle and github dataset accounting to total 1300 COVID19 and 1200 normal chest x-rays. For breast cancer we used kaggle dataset. For symptoms and disease we used a csv file for NLP training.
How we built it
Challenges we ran into
Training using wit.ai was difficult and between the event there were some changes made to wit.ai. Integrating tensorflow.js models with messenger webhook and integrating all of it into one single project was challenging.
We are trying to make our bot perfect and will research methods of implementation that could improve the accuracy of our bot. We will experiment with other architectures for training our model to improve efficiency. This can be achieved approximately in a time span of a month. After this is done we may look for funding and make it available to people.
Accomplishments that we're proud of
What we learned
Deeper understanding of Facebook Messenger architecture and how wit.ai works. Training of NLP using wit.ai. Machine learning model creation, conversion to tensorflow.js and integrating it with messenger
What's next for RoboDoc
At present we have added 21 diseases we will be expanding it to 87 diseases for predictions using all symptoms.We are trying to make our bot accurate and as it is used more we will train it for more symptoms and diseases. We will research methods of implementation that could improve the accuracy of our bot. We will experiment with other architectures for training our model to improve efficiency. We will be including more medical models for diagnosis of more diseases using x-rays and MRI's.This can be achieved approximately in a time span of a month. After this is done we may look for funding and make it available to people.