Bye Bye Covid

Using Machine learning to revolutionize COVID-19 diagnosis

Comment

Inspiration

Bye Bye COVID aims to take off the burden off our healthcare system in case of a potential 4th wave of COVID-19. SARS-COV2 has taken an estimated 4.9 million lives worldwide. With surges, emerging variants, and limited access to vaccine and testing materials, experts agree that the end is not in close reach just yet. In fact, our lives as we know it may have changed indefinitely. Hospitals are forced to cancel surgeries and reserve entire floors for COVID-19 patients, and ICUs are at full capacity across the globe. Today, COVID-19-induced pneumonia is raging through critical care units, and Healthcare providers are facing an immense feeling of burn-out - we aim to ease that burden through Artificial Intelligence based technology.

What it does

The goal of our hack is to ease the process of recognizing COVID-19 induced pneumonia, especially in situations where there are time constraints and limited resources for testing - these can quickly become life-or-death moments. Our model inputs x-ray imaging of the chest and outputs the probability that the result is: A) COVID-19-induced pneumonia, B) Normal or C) Other Pneumonia. Our 2nd model inputs cough audio and outputs the probability that the result is: A) COVID-19 B) NOT COVID-19

How we built it

  • Chest X-ray COVID Classifier - Used publically available dataset. Cleaned and preprocessed it. Applied data augmentation (randomly scaling, shear zoomed, rotated, and flipped the training set) to prepare for real-world situations. Then designed a CNN to do a multi-class classification

  • Cough Sound COVID Classifier - Used publically available dataset, then did feature extracting on the sound files using librosa library after cleaning the dataset and processing it. Used the training and testing set created from the new features, made machine learning models.

Challenges we ran into

  • Computation power (Deep learning takes a lot of processing power)
  • Lack of data (especially for cough samples)
  • Group made quitting, leaving me alone to complete

Accomplishments that we're proud of

  • Models that can classify sound and X-rays to tell users if they have COVID-19
  • Finished Pitt Challenge

What we learned

  • Uncovered the complexities of COVID-19-induced pneumonia, which opened our eyes to how pressing this matter is.
  • Enhanced data cleaning and processing skillset
  • Deep learning

What's next for Bye Bye Covid

  • Train our model with more datasets and for longer epochs
  • Create a user interface with sleek design
  • Highlight suspected COVID-19 pneumonia markers on X-Ray image

Built With

Share this project:

Updates