Learning to localise medical anomalies from multiple synthetic tasks

Self-supervised anomaly detection involves training models to identify synthetic anomalies introduced into otherwise normal data.
We use multiple synthetic tasks for cross-validation, early-stopping to avoid overfitting. We readily outperform state-of-the-art methods on both brain MRI and chest X-rays. We investigate how model performance varies as
different numbers of task are allocated to training versus validation.
Baugh M et al.: nnOOD: A Framework for Benchmarking Self-supervised Anomaly Localisation Methods. In
UNSURE 2022, Held in Conjunction with MICCAI 2022, Singapore, September 18, 2022, Proceedings
2022 Sep 14 (pp. 103-112).


General research area Medical Imaging