Training

Through MaxBI, the MPI for Medical Research in Heidelberg gave me the opportunity to participate in a three day long image analysis workshop hosted at their institute. The well structured recap of the basics of image analysis that started off that course will definitely help me to teach the users of the imaging platform at our institute. I have referred to the Neubias website (https://neubias.github.io/training-resources/all-modules/) that provides resources to learn, understand and train all kinds of basic analysis techniques many times since then. However, we were also given practical examples of how to leverage pixel classifiers and got comprehensive advice of how to solve very specific image analysis problems.

The two major learnings I take away from this course however, are the more extensive use of distance maps in all kinds of image segmentation and analysis workflows as well as a toolkit of how to provide my users with interactive FIJI macros with GUI that we now already use to standardize and speed up basic workflows at our institute, while giving scientists the flexibility to change parameters on the fly according to their input data and needs. Thanks again for this opportunity, the MaxBI network that made me aware of the offer and the financial support for covering traveling costs.

- Kenny Mattonet, Max Planck Institute for Heart and Lung Research -

From the 6^th to the 9^th of November I attended the workshop “developing community standards for bioimage analysis”. We were a group of 11 expert bioimage analysts from UK, USA, Israel, Germany, Brazil, Switzerland, and Finnland. During our time in Heidelberg we worked on:

1. Define a curriculum for teaching bioimage analysis and address
   1. Important topics/concepts to teach.
   2. Order to teach these topics. This automatically gives a beginner, intermediate, advanced level.
   3. Some topics can have multiple level of complexity. Define those levels.
   4. Define exercises to best illustrate concepts.
2. Set up a repository for image data material. Ideally, images are associated with exercises and examples connected to the curriculum.
3. Work on best practice to teach and motivate students
   1. Lecture slides, videos, online interactive material, … What works best when, what is available.
   2. Define questionnaire to assess the knowledge level of participants. These could be exams questions or an objective way to set the level of a course.
   3. How many concepts can be digested within a certain period of time?
   4. How to include students own problems in a course?

The meeting ended up with plenty of ideas and follow up online meetings where we further discuss. To point 1, to get a feedback from a larger community, we are working on a first release of a curriculum draft on https://forum.image.sc/. To point 2, we reached out to EBI and the bioImage archive to setup a database for example and openly accessible images. Finally, to point 3, we would like to organize a course with the software carpentry on teach the trainer.

- Antonio Politi, Max Planck Institute for Multidisciplinary Sciences -

Researchers at our institute frequently analyze crowded lung and heart tissue samples in 3D and time lapses. Unfortunately, it is notoriously difficult to segment such samples with sufficient quality using these conventional image-processing workflows and it has turned image analysis into a major bottleneck in our facility. To date, specialized deep learning algorithms like Stardist and Cellpose are the best tools we have found to help our users get the high-quality data they need in a timely, efficient manner. So, I wanted to learn more about it.
The EMBL course “Deep learning for image analysis” provided me with the perfect opportunity to gain an overview over the available tools. It also showed me how to work with them, when to use them, and, most importantly, how to check the validity of the results. It was an intense learning experience and everything about it was geared towards creating the best possible learning environment: the location, the structure of the course, and the team of highly skilled trainers enabled us to train accurate and effective custom models in just a few days.
In the few weeks since the course ended, we already implemented the new knowledge from the workshop in our facility. We automated the analyses for three different projects that, lacking a better alternative, had to be manually quantified before. If needed, we can create simple custom solutions for future projects any time and several deep learning tools for image analysis that are now deployed on our workstations.
The EMBL workshop was a valuable learning opportunity for our imaging platform, and added many new tools to our repertoire that already benefit the facility and our researchers. I deeply thank the MaxBI network for making this possible by covering parts of the travel expenses.

- Kenny Mattonet, Max Planck Institute for Heart and Lung Research -