Settings
  • Light
  • Dark
  • Auto
Select Page

In the second episode of ESR Connect’s weekly show on artificial intelligence, Reasons to do AI with Friends, Prof. Helmut Prosch, from the Medical University of Vienna, presented the case of a 76-year-old male patient with a chronic cough who used to smoke heavily for 50 years. CT scans showed signs of lung fibrosis, in which honeycombing was visible. Prosch followed this by presenting four different disease patterns for fibrosis.

Prof. Georg Langs, also from the Medical University of Vienna and who you may remember from the first episode, presented two ideas for machine learning solutions that can help to detect fibrosis. The first idea was supervised training of a model in which an algorithm is taught to classify patterns – like honeycombing or ground glass – in new image data. In contrast, in unsupervised machine learning, the training model is not asked to repeat what was done in the training set but is provided with several observations and lets the model find structure in the data. Thus, the algorithm identifies patterns linked to the disease by clustering. The requirements for such an algorithm are enormous amounts of imaging data and, ideally, a segmentation of the lung; the more diverse the data set, the better, and it should consist of multiple modalities, centres and populations. Subsequently, Prof. Langs provided insight into the workflow for a project like this in both a supervised and unsupervised training set; Langs went on to explain the advantages and limitations of these options. The practical use in lung screening was also covered in detail and he explained why radiologists actually profit from such machine learning tools. Langs also described how the results can then be validated, especially clinically.

This week, Theo’s Technical Top Tips focused on data anonymisation. Theo Barfoot is an MRI physicist at the Royal Marsden London and presents one valuable detail per episode. In the second episode, he explained what needs to be considered when transferring patient information from a clinical environment to a scientific environment to ensure patients’ privacy without losing information necessary for the research.

A highlight of this week’s episode was “Our most painful mistakes”, and not only because the hosts were using a loud buzzer to underline them, but also because they openly shared five pitfalls during the project, which was the audience loved.

As there were numerous questions and comments in the live chat throughout the show, Prof. Langs and Prof. Prosch went above and beyond, spending extra time at the end of last night’s episode to ensure that all questions were answered.

One viewer wrote, “Georg & Helmut, you are GREAT!” and the editors agree! This new format really does stand out as it presents state-of-the-art content that is both entertaining and informative.

This season, Reasons to do AI with Friends will run live every Wednesday night at 19:30 CET until February 19 and present use cases for various medical issues. The next episode, “The one with the sixpack”, will be hosted by Andrea Rockall, Shah Islam and Fahdi Kanavati, from the Imperial College London, on January 29 at 19:30 CET. Registration is free of charge.

All episodes will be available on-demand for purchase on connect.myesr.org.

Read more on Episode 1, “The one where we classify breast cancer with MRI” here.
Read more on Episode 3, “The one with the sixpack” here.
Read more on Episode 4, “The one that might be renal cancer” here.
Read more on Episode 5, “The one where we find the connection” here.
Read more on Episode 6, “The one with whole body MRI” here.

Want to deepen your AI knowledge? Check out the reading recommendations from our experts here.

Latest posts