Inspiration
Pneumonia is a prevalent disease that infects millions of people a year. We had come across an article in one of our classes at school and learned that many pneumonia cases are often misdiagnosed as other less serious medical conditions, such as the common cold, or overdiagnosed, forcing patients to go through unnecessary treatment that puts a strain on their immune systems. We wanted to create a solution to this problem that could help developing countries where lack of good education and physicians are present.
What it does
We created a platform that informs the user of their status regarding the infectious disease. The website allows the user to upload a picture of their x-ray, and the CNN (convolutional neural network) scans it and predicts whether the scan shows indications of pneumonia based on a dataset from kaggle.
How we built it
First, we created a trello board to stay organized within the tasks we had to complete. One of us began on the front end by using sublime text to create a website using HTML, whilst the other worked on the CNN model using jupyter notebook. Next, we debugged and improved the accuracy of the model together and proceeded to merge frontend and backend using Flask. Lastly, we tested our complete product.
Challenges we ran into
We faced multiple challenges throughout the development of our site. First, when creating the CNN model, the path to the dataset was incorrect, so our model wasn’t getting trained. We figured that out pretty quickly, though, but the next issue we faced was related to overfitting the model. To combat this issue we used early stopping and a learning rate scheduler.
Accomplishments that we're proud of
We’re really proud that we were able to make a CNN so efficiently today with a high 70s rate of accuracy. We also really like the fact that we made an interactive and aesthetic web page using HTML, CSS, and Bootstrap that highlighted our problem.
What we learned
We learned how to work with CNNs and Flask. Additionally, debugging was a great skill that we developed here (and the help the mentors provided was very helpful when we weren’t able to figure out what the issue was).
What's next for Pneumonia Detector
For our Pneumonia detector some next steps include finding a larger dataset to improve the accuracy of the model and creating models for other infectious diseases that use scans and adding them to our website, an example would be bronchitis.
Log in or sign up for Devpost to join the conversation.