The BENDER video series is available on the BeLearn platform
The premise is that of a new graduate student Satish, who is supported by a senior graduate student, Mike, through various steps in the process of building a state-of-the-art model to solve a medical image analysis problem. We start with a checklist of what to do when one receives clinical data, dealing with issues like data curation, anonymisation, exploratory statistics, etc. Then we move on to a short discussion on clinical terminology – how it is important for all the stakeholders in these interdisciplinary projects to communicate effectively. Next, we cover some tips and tricks around actually building a model: how one can go from a naive implementation to a state-of-the-art benchmark beating model on a publicly available data set. This also includes some practical advice on tools to use. Finally, we cover some under-appreciated topics like external test set evaluation, clinical relevance and robustness, and also bonus content on dealing with ‘reviewer 2’.
This short video series is complemented with a more thorough and updatable GitHub repository where we include more details, references, useful links and code snippets, where we invite the community to contribute.
We hope this would be a great learning companion for a new student venturing into this field, with everything covered in a fun and engaging manner.
Our upcoming season for BENDER is in preparation and the team is excited about the new topics and feedback from the community!