Project Name: AI Radiologist
We wanted to help with a very relevant, wide-spread, and unsolved problem in today’s world. COVID-19 infected over 95 million people and has affected the lives of everyone across the world. We felt obligated to assist in dealing with this unprecedented threat to humanity.
What it does
Our project uses convolutional neural networks (CNNs) to distinguish between common pneumonia, covid-induced pneumonia, and an uninfected person through the deep analysis of lung x-rays. It can distinguish between the different illnesses at over a 95% accuracy rate.
Challenges we ran into
Finding a large amount of data was by far the hardest part. We looked through dozens of databases and manually-curated datasets to the point where we were going to create a dataset by ourselves. Finally, we stumbled upon an existing one on Kaggle out of sheer luck.
Accomplishments that we're proud of
Our AI easily outperforms most radiologists! The accuracy of radiologists in detecting COVID-19 in an x-ray is between 76%-81%. This proves how AI can greatly help improve patient care.
What's next for AI Radiologist
We will create a deeper and more robust neural network to detect COVID-19/pneumonia in x-ray scans with higher accuracy and more speed. We could expand to more than just chest x-rays in the future.
Python, Tensorflow, Kaggle, CUDA/NVIDIA