1 of 9

ECG AUTOMATED INTERPRETATION

AZIM SHAFI

JULIAN ROBLEDO

SENG JHING GOH

NICANGEL SANCHEZ

​

2 of 9

GOAL & MOTIVATION

Goal

    • Produce an autonomous system
      • Take in dataset
        • ECG graphs
      • Interpret the graphs
        • Determine the condition of the patients

Motivation

    • ECGs not included in the physical exam of student athletes
    • Lead to treatable heart abnormalities going ignored

3 of 9

KEY FEATURES

CSV File & Database

    • CSV file:
      • The user will be able to:
        • Take multiple CSV files and move them into one large file
        • Modify the data from the CSV file
        • Upload the CSV file into the database
    • Database
      • Composed of patients' names, age, ethnicity, gender, BMI, diagnosis, link to pdf file, extracted image of ECG graph
      • Users will be extract what kind of information they want and manipulate it

PDF Data Clean Up

    • Automated approach to remove all identifying patient info
    • Clean the ECG graph prior to extracting any data from the PDF
    • Users will be able to select a dataset that contains the PDF files
    • The software exclude the patient’s identifying info autonomously without further intervention

​

Extracting ECG graph from the PDF Files

    • Extract an ECG graph that will be automated depending on the type of ECG graph by selecting a section of the PDF file using cartesian coordinates
    • Crop out an image of the ECG graph

ECG Interpretation

    • Software will automatically interpret the ECG graph and determine whether it has any abnormalities
      • Take account the patient’s relevant physical characteristics

4 of 9

NOVEL FEATURES/�FUNCTIONALITIES

  • ECG Automated Interpretation
    • Using a Convolutional Neural Network (CNN) to identify abnormalities in an electrocardiogram
    • With a dataset of 40,000 ECG’s
      • Hope to be able to train the neural network to obtain high accuracy for the identification, verification of the diagnosis

5 of 9

TECHNICAL CHALLENGES

Extracting/Cleaning PDFs

    • The ECG machines used by the client output the graphs as PDFs
      • Work with that format
      • One of us has ever worked with it

01

Implementing/Training the CNN

    • Client wants the ECG interpretation to be done using a CNN
      • No experience with neural network or machine learning

02

Creating/Managing the Database

    • The need to store information for each ECG
    • Less to do with working with databases but more towards managing a large database

03

6 of 9

MILESTONE 1

  • Compare and select technical tools:
    • Database consisting of one CSV file made from multiple CSV files or Database with multiple CSV files.
    • Tools to work with CSV files: Apache Spark, or Python
    • Output image format PNG and JPG
    • Tools for cleaning up and extracting PDF files
  • Provide demos:
    • Modified a dataset from the CSV file using Spark and Python 
    • Working database of 2 databases for comparison
    • Extracting image and cleaning PDF file
    • PyTorch demo in the teams development environment
  • Resolve technical challenges
    • How to read and extract an image from a PDF
    • How to “clean” an image
    • How to manage a database/selecting what database is best
  • Create Requirement Document, Design Document, Test Plan

7 of 9

MILESTONE 1

  • Collaboration Tools
    • Communication: Discord, Slack, Text Group chat
    • Document Sharing/Presentation: Google Drive, Github
    • Version Control Environment: Github, Gitlab
    • Task tracker: Github Projects, Monday, Jira

8 of 9

MILESTONE 2

  • Implement, test, and demo:
    • Script that extracts image/data from PDF and stores it in our database

9 of 9

MILESTONE 3

  • Implement, test, and demo:
    • Script that cleans extracted image/data
    • PyTorch in working environment

​

  • Implement, test, and demo:
    • Script that cleans extracted image/data
    • PyTorch in working environment

​