DAIM: Images Closing Seminar

Let’s Do Digital Team

Closing Session

  • Welcome to the closing seminar of the Data Analytics in Medicine - Images course.

  • What did you learn?

  • What are your immediate thoughts?

Refresher Core Learning outcomes

  • Understand how computers process image.
  • Understand how computers learn
  • Diving deep into DICM vs. other image formats
  • Understand types of machine learning
  • Understand the trade-off’s in machine learning / AI models.

What we have covered

  • Module 1 - Overview
  • Here we covered the overall structure of the course and introduced the tools we would be using

Discussion

  • What worked well in this unit?
  • What would you suggest to improve this unit?

What we have covered

  • Module 2 - Introduction to images
  • Here we covered:
    • The basics of how images are represented digitally
    • Common image operations such as convolution

Discussion

  • What worked well in this unit?
  • Was the workshop pitched at an appropriate level of difficulty?
  • What would you suggest to improve this unit?

What we have covered

  • Module 3 - DICOM
  • Here we covered:
    • The basics of how medical images are represented digitally
    • The metadata that is associated with medical images, and how to extract it

Discussion

  • What worked well in this unit?
  • Was the workshop pitched at an appropriate level of difficulty?
  • What would you suggest to improve this unit?

What we have covered

  • Module 4 - AI
  • Here we covered:
    • The basics of neural networks
    • How to build and evaluate a binary classifier
    • Clinical considerations for medical AI systems

Discussion

  • What worked well in this unit?
  • Was the workshop pitched at an appropriate level of difficulty?
  • What would you suggest to improve this unit?

Extending each module

Further material for module 2

  • TODO

Further material for module 3

  • TODO

Further material for module 4

  • Try adapting a different model to the same dataset
  • Kaggle runs competitions for machine learning tasks
    • There is a wealth of openly-available datasets to work with on this site.
  • Further examples could include:

Going Forwards

The Recipe

  • Coding skills : You need not just ability to code but ability to troubleshoot, learn multiple languages. e.g., SQL, C#, Rust or improved fluency in Python
  • Theory: PhD in Maths is not necessary (can be helpful to an extent!) but need to understand it’s all matrix operations.
  • Project Management Skills: Time management, customer management, agile working skilsl, communication skills
  • Clinical Translation: You either possess domain knowledge or you work with people with domina knowledge

My personal opinion

  • You need all 4 of them. You may want to specialise into one of them
  • You may want to do most of the coding -> ML engineer.
  • Theory and Building models -> ML practitioner / data scientist
  • Project Mananger: might not even need coding skills
  • Clinical Translation: might be a doctor who understand how AI works and help procure right solution for NHS.

What do you want to do next?

  • Machine Learning Engineer : perhaps give yourself dedicated time to learn more theory & start building, will require data science x Engineering skills.
  • What do ML Engs do: translate real world data and real world models to “production”
  • Skills: Fluency in Python, SQL, Docker Deployment, Networking, Optimising

What do you want to do next?

  • Machine Learning Practitioner : find an area of interest, join a research team.
  • What do ML Practitioners do: may involve in data cleaning, building models, help visualise
  • Skills: may chose to specialise into structured data (mainly tabular data) or unstructured data (images) or natural language processing (NLP)

What do you want to do next?

  • Clinician with AI knowledge :
    • Read more. Find flaws.
    • Think how could this AI model fail and how it could fool you.

Getting involved!

Learn more!

  • Coursera, Udemy, Youtube, Codeacademy, Datacamp…
  • Books
  • ChatGPT I admit I use ChatGPT but ChatGPT doesnt improve your skill. There is no shortcut! :)

Books

  • Structured Data: ISLR, Sklearn, Statistics for Data Science
  • Image Data: Recommend Pytorch
  • Deep Learning with Python by Francois Chollet
  • Mastering Pytorch by Ashish Ranjan Jha
  • Deep Learning with PyTorch by Eli Stevens
  • Fluent Python by Luciano Ramalho

Build, Build, Build!

We know it is hard to do alongside medicine / job. But you got this.

Stay in touch!

It is a friendly and exciting community.

Thank you

  • Please let us know if you have any further feedback or questions.