Computer Vision News - April 2023

44 Congrats, Doctor Camila! Despite mounting evidence that data drift causes deep learning models to deteriorate over time, the majority of medical imaging research is developed for - and evaluated on - static close-world environments. There have been exciting advances in the automatic detection and segmentation of diagnostically-relevant findings. Yet the few studies that validate their performance in real clinics are met with disappointing results and little utility as perceived by healthcare professionals. This is largely due to the many factors that introduce shifts in the data distribution, from changes in the acquisition practices to naturally occurring variations in the patient population. If we truly wish to leverage deep learning technologies in the healthcare sector, we must move away from close-world assumptions and start designing systems for the dynamic clinical world . Detecting a deterioration in model performance This entails, first, the establishment of reliable quality assurance mechanisms with methods from the fields of uncertainty estimation, out-of-distribution detection, and domain-aware prediction appraisal. We propose two approaches that identify outliers by monitoring a self- supervised objective or by quantifying the distance to training samples in a low-dimensional latent space. We also explore how to maximize the diversity among members of a deep ensemble for improved calibration and robustness; and present a lightweight method to detect low-quality masks using domain knowledge. Camila González recently completed her PhD at the Technical University of Darmstadt, Germany. Her research mainly focused on adapting deep learning methods for dynamic clinical environments. She plans to continue her research as a postdoctoral scholar at Stanford University, in the Computational Neuroimage Science lab. Congrats, Doctor Camila! OOD detection Prediction coherence Uncertainty Model performance Uncertainty estimation OOD ID