MOOD 2020: A Public Benchmark for Out-of-Distribution Detection and Localization on Medical Images

TLDR

Detecting out‑of‑distribution data is a major challenge for safe deployment of medical machine‑learning models, as unseen cases can trigger incorrect, over‑confident predictions, and out‑of‑distribution detection aims to preempt such errors and aid clinicians, yet no open benchmark existed for medical imaging. The authors introduce the Medical‑Out‑Of‑Distribution‑Analysis‑Challenge (MOOD) as an open, fair, and unbiased benchmark for out‑of‑distribution methods in the medical imaging domain. MOOD supplies a standardized dataset and evaluation protocol that allows researchers to assess and compare OoD detection algorithms on medical images. Analysis of submitted algorithms shows that performance correlates strongly with perceived difficulty and varies widely across anomalies, making clinical recommendation difficult, while ranking also correlates with performance on a toy test set, indicating the benchmark could serve as a useful proxy during development.

Abstract

Detecting Out-of-Distribution (OoD) data is one of the greatest challenges in safe and robust deployment of machine learning algorithms in medicine. When the algorithms encounter cases that deviate from the distribution of the training data, they often produce incorrect and over-confident predictions. OoD detection algorithms aim to catch erroneous predictions in advance by analysing the data distribution and detecting potential instances of failure. Moreover, flagging OoD cases may support human readers in identifying incidental findings. Due to the increased interest in OoD algorithms, benchmarks for different domains have recently been established. In the medical imaging domain, for which reliable predictions are often essential, an open benchmark has been missing. We introduce the Medical-Out-Of-Distribution-Analysis-Challenge (MOOD) as an open, fair, and unbiased benchmark for OoD methods in the medical imaging domain. The analysis of the submitted algorithms shows that performance has a strong positive correlation with the perceived difficulty, and that all algorithms show a high variance for different anomalies, making it yet hard to recommend them for clinical practice. We also see a strong correlation between challenge ranking and performance on a simple toy test set, indicating that this might be a valuable addition as a proxy dataset during anomaly detection algorithm development.