ML Diagnostics

Build ML Application allowing users to fetch their clinical risk by parsing basic clinical attributes. The application generates a Clinical Risk Score based on Probabilistic Models.

Comment

Inspiration

  • The annual median total medical costs for heart failure care are estimated at $24,383 per patient in United States translating to $351 Billion per year – CDC

  • 99% of hypertension and stroke instances were detected by a simple screening using the model – McKinsey

  • ECG cost ranges between 76-240$ and CT Scan ranges from 300$ to 6750$ in US – HHS.gov

  • 70% of global severe cardiovascular disease casualties occur in low and middle income countries – WHO

  • 23M Casualties forecasted per year in low-income countries by 2030 – World Heart Federation - 2019

  • 2.2 Million hospitalizations due to heart strokes resulting from ignored previous symptoms – CDC 2018

What it does

Our Application Generates a clinincal risk associated with heart disease. Users can keep a track of their historical records and analyse it.

How we built it

  • Database:
    • AWS DynamoDB
  • Front End
    • React
    • Next.js
    • JavaScript
    • CSS
  • Back End
    • Node.js
    • Express.js
    • TypeScript
    • Python
  • Model Development
    • Python
    • Sklearn - GBM Classifier

Challenges we ran into

  • Dependency issues
  • Security patch issues with AWS while using public github repository
  • Integration of Python with Node.js
  • React Table dev bugs

Accomplishments that we're proud of

  • Yayy our application is running smoothly
  • We are able to retrieve historical results and a smooth connection between AWS services and local host

What we learned

  • Python Integration with Node.js
  • Handling dependency issues quickly by using debugging tools and console
  • Virtual Environment and its importance in building a stand alone application

What's next for ML Diagnostics

  • Since our approach is scalable, we plan to build multiple ML models targeting major diseases impacting high risk population like pulmonary diseases
  • We will host our application in AWS EC2 and use S3 Buckets for storing data for different medical Facilities
  • Use Spark Streaming Application for making it a highly responsive application with very low runtime for complex Mllib models
  • Interactive visualizations and insights generating module using user medical information
Share this project:

Updates