Inspiration
The motivation for our project stems from an inefficient testing system. At centers, patients have to wait in long lines, book appointments much in advance, and may potentially spread the disease until their results are retrieved. Our project gives a pre-diagnostic scan for patients who are in consideration of testing or may be experiencing symptoms.
What it does
Our solution to the limited types of Covid 19 tests is a deep learning algorithm that uses convolutional neural networks to process chest x-ray images of patients. The website and mobile app solutions are instantaneous, automated, and not subject to human error, which is an issue with conventional covid tests because people have to do them manually, leading to high misdiagnostic potential.
How we built it
Our code has some unique characteristics that allow it to have high accuracy. We used a multitude of python libraries such as the pandas and numpy libraries to help our process run smoother. We trained it on a VGG-19 convolutional neural network which goes through a number of hidden layers to ensure the model is trained well. Our data comes from radiologists in global labs and is divided into 80% training data and 20% for testing purposes once the model has been trained.
Challenges we ran into
Time for training the model - took significant time to train over 300 images
Accomplishments that we're proud of
Creating a model with a 94% success rate!
What we learned
Deep-Learning algorithms need tons of training data to be successful and highly accurate Patient data is hard to receive since it is oftentimes highly classified data
What's next for Covid-19 Diagnostic Tool
We plan to transition to Apple Watch reminder apps and to expand these scans to pneumonia etc. We also plan to create a tool to connect doctors with patients with the scan
Built With
- as
- html
- java
- machine
- machine-learning
- python
Log in or sign up for Devpost to join the conversation.