FrOoDo: Robust artefact detection and image quality control in AI-driven computational pathology

Poster presented as part of the Crick BioImage Analysis Symposium.

As rapidly developing deep learning methods minimize laborious image  processing, they are able to significantly speed up the diagnosis  prediction in the field of computational pathology. However, for  accurate classification or precise segmentation a robust model is  crucial. Often deep learning models show excellent results from one  laboratory setting, however minor differences in staining protocols or  image acquisition can lead to specific artefacts. Such artefacts fall  into out-of-distribution data range and thereby tangle the automated  deep learning image processing causing confident failures. To address  this problem, we have developed a framework for out-of-distribution  detection (FrOoDo) for detecting such artefacts in an ad-hoc fashion.  Furthermore, we added advanced artefacts simulations to improve our  frameworks reliability and showcase the necessity for prediction quality  control. Our results show that while all artefacts have an increasingly  adverse effect on the prediction performance w.r.t their severity.  FrOoDo is perfect for reliable detection of more severe artefacts, thus  allowing for the more accurate disease prediction. As our framework  builds a foundation for quality control for in computational pathology,  we want to extend FrOoDo to biomedical images and explore more  applications in the future. 

Code is available at:  https://github.com/MECLabTUDA/FrOoDo.

Permission has been given by authors to upload to Crick Figshare. Copyright remains with the original holders.