During the master's, we had a course called Intelligent Systems in Medical Imaging. In this course we got tought about the basics of medical imaging and state-of-the-art techniques of gaining insights from the data. We were tasked to compete on The Grand Challenge, "A platform for end-to-end development of machine learning solutions in biomedical imaging". Here, competitions are hosted where they provide a medical imaging dataset.
The challenge our team chose was to predict the head circumference based on ultrasound images (hopefully explains the header image). The leaderboard is open across the world, and this challenge had more than 2,000 entries at the time. After consideration we opted for the nn-Unet architecture. This architecture is an interesting approach to machine learning in itself, makes me think of the Bitter Lesson by Richard Sutton (good read!).
We ran the algorithm on a compute cluster provided by the university, where we learnt to work with Slurm and it was one of our first interactions with a partition on a linux machine.
In the end, we got first place out of 2,212 entries at the time. And at the time of writing, we are still at number one! The score was a mean absolute error measure where we achieved 0.74 mm mean and 0.73 mm standard deviation. See the challenge details here.