Enhancing reconstruction-based out-of-distribution detection in brain MRI with model and metric ensembles

Background and Objective:

Out-of-distribution (OOD) detection is crucial for safely deploying automated medical image analysis systems, as abnormal patterns in images could hamper their performance. However, OOD detection in medical imaging remains an open challenge. In this study, we aim to optimize a reconstruction-based autoencoder specifically for OOD detection. We address three gaps: the underexplored potential of a simple OOD detection model, the lack of optimization of deep learning strategies specifically for OOD detection, and the selection of appropriate reconstruction metrics.

Methods:

We investigated the effectiveness of a reconstruction-based autoencoder for unsupervised detection of synthetic local and global artifacts in brain MRI. We evaluated the general reconstruction capability of the model, analyzed the impact of the selected training epoch and reconstruction metrics, assessed the potential of model and/or metric ensembles, and tested the model on a dataset containing a diverse range of artifacts.

Results:

Among the metrics assessed, the learned perceptual image patch similarity (LPIPS) and the contrast component of structural similarity index measure (SSIM) consistently outperformed others in detecting homogeneous circular anomalies. By combining two well-converged models and using LPIPS and contrast as reconstruction metrics, we achieved a pixel-level area under the Precision–Recall curve of 0.66. Furthermore, with the more realistic OOD dataset, we observed that the detection performance varied between artifact types; local artifacts were more difficult to detect, while global artifacts showed better detection results.

Conclusions:

Our study shows that a reconstruction-based autoencoder, when combined with appropriate metrics, enhances OOD detection in brain MRI. These findings emphasize the importance of carefully selecting metrics and model configurations and highlight the need for tailored approaches, as standard deep learning approaches do not always align with the unique challenges of OOD detection. Improving OOD detection can increase the reliability of automated medical image analysis.