ETHZ

Machine Learning for Health Care course at ETH Zurich

Over the past years, more and more machine learning-powered technologies have been integrated into the health care sector. This course shows how machine learning can be used to solve problems in biomedicine. Concrete problems such as data privacy and interpretability of results are also discussed and possible solutions are presented.

This course contains a normal lecture and only tutorials during the first few weeks. What’s special is that there are no theory exercises, so you have to decide about what’s important for the exam. Another specialty is that students have the opportunity to present current research work. This is mainly because the subject area is developing rapidly and the course can thus keep pace with research in terms of content.

A general explanation of each part and what other courses I have attended can be found here.

Paper presentations

The paper presentations are about provided papers in various areas of health care, such as genomics, medical images, and more.
Each presentation is expected to be about 25 minutes long and can be given by up to two students.

Projects

For each project, you have about 20 days’ to complete and each project team should consist of three people. The projects are about topics that are presented during the lectures, i.e. at the beginning of the course when time-series data and how to deal with it is presented, then the project is about time-series.

The four projects are:

  • ECG time series classification (multi-class)
  • NLP task (two tasks)
    • predict readmission within 30 days of discharge, based on ordinal, categorical and text features
    • participate in the COVID-19 Open Research Dataset Challenge
  • Image segmentation
    • on 3D prostate magnetic resonance images (MRI)
  • Splice site classification

Grading

Everyone has to work on three out of four projects, but if you make a paper presentation, that presentation can replace one project.

The average of the best three projects or the best two projects and the paper presentation then contribute 30 % to the final grade, the remaining 70 % is based on the written final exam.

Topics

The following topics are covered:

  • Sequence Analysis and Time Series
  • Survival Analysis
  • Natural Language Processing of Clinical Text
  • Representation Learning
  • Ethics and Big Data
  • Privacy Preserving Computing
  • Medical Imaging Analysis
  • Interpretability of Machine Learning Models
  • Supervised Methods for Genetics and Transcriptomics
  • Unsupervised Methods for Genetics and Transcriptomics