The inspiration behind our latest project came from a deep desire to improve the accessibility of medical care for people living in rural areas. As developers from third-world countries ourselves, we understood all too well the struggles faced by those who lack access to proper medical facilities.
Our team wanted to create a solution that would bridge the gap between these individuals and medical professionals, so we set out to create an app that would allow users to receive medical advice and diagnoses using machine learning models.
By inputting their symptoms into the app, users can receive predictions for potential diseases and conditions, allowing them to make informed decisions about seeking medical treatment. We believe that this app has the potential to be particularly beneficial for those in rural areas who may not have easy access to medical doctors or facilities.
For us, this project was an opportunity to use our skills and expertise to make a real difference in the lives of people in our own communities and beyond. We're excited to continue developing and improving this app and to see it make a positive impact on the world.
What it does
MEDDIBIA is a personal AI health assistant that puts greater control over health in the hands of users living in rural communities. With MEDDIBIA, users can describe their symptoms to a chat assistant and receive a likely diagnosis, along with more information about their diagnosis and symptoms. Additionally, MEDDIBIA enables users to get diagnoses for skin conditions and aberrations by simply taking a picture of the affected area. This feature is especially important for those with limited access to healthcare professionals or specialized facilities. By providing personalized care and making it easier to manage health conditions, MEDDIBIA empowers users to take control of their health and improve their quality of life.
How we built it
Our team utilized cutting-edge machine learning techniques to develop an innovative solution for identifying and diagnosing skin conditions and diseases. To accurately identify skin conditions, we experimented with various pre-trained models, including VGG16 and EfficientNet, to extract features from images from the dermnet dataset. We then trained and evaluated deep neural network classifiers, ultimately selecting a model with approximately 70% accuracy. For symptom identification, we employed GPT-3, a state-of-the-art language model, to preprocess natural language input from users into symptom labels, which served as input to our machine-learning model. This approach resulted in about 87% accuracy in predicting disease labels. To further assist users, we used GPT-3 to provide helpful descriptions of the predicted disease. Our app's backend was built using Flask API and deployed on Heroku, while the cross-platform frontend was developed using Flutter, making our app easily accessible to users across multiple devices.
Challenges we ran into
The construction of MEDDIBIA was an interesting and challenging task. The first problem we encountered was locating suitable datasets for our machine learning algorithms. We needed to obtain a dataset with over 40 diseases and appropriately identify them using their symptoms for our disease classification algorithm. To maximize machine learning, we needed to obtain a dataset with rich photos for our skin disease detection model. The next problem was to discover effective machine learning techniques to use with our dataser to produce modesl. To acquire accurate findings, we needed to determine the machine learning technique that performed best with our dataset. Another difficulty we encountered was integrating our machine models to our mobile application. Creating a machine learning model is one thing, but we also needed to guarantee that our model was user-friendly and easily assessable via our application. Constructing MEDDIBIA was difficult, but we were able to overcome the obstacles that the journey posed in order to complete our project, MEDDIBIA.
Accomplishments that we're proud of
We are proud of the knowledge we gained from this experience. Building MEDDIBIA was challenging and the process allowed us gain useful knowledge on several technologies. In addition. We are proud of being able to create such a rich project within 24 hours. Iin 24 hours we were able to create a project that uses machine learning to predict diseases users might have using their symptoms as values.
What we learned
We learned how to integrate flask APIs to a Flutter application while developing MEDDIBIA. Our application is built on Flask, which connects machine learning models to a Flutter application. We discovered how to leverage openAI APIs. Our application makes advantage of openAI APIs to fine-tune user input before passing it to our model. We also learned about the multinomial naive bayes machine learning method, Flutter, and Flask.
What's next for MEDDIBIA
Looking to the future, we plan to suggest nearby hospitals and clinics, potential treatments and connect patients directly with healthcare providers through MEDDIBIA. These features will personalize care, streamline the process of accessing medical assistance, and provide the latest treatment options. We are excited to continue our work in revolutionizing healthcare and improving health outcomes for our users.
Log in or sign up for Devpost to join the conversation.