Projects

Automatic diffusion models for anomaly detection.

Reversing the Abnormal: Pseudo-Healthy Generative Networks for Anomaly Detection.

Novel deformable auto-encoders for unsupervised outlier detection.

UAD faces significant challenges tied to biases from different sources, including scanners, sex, and race. Future UAD models should maintain sensitivity to pathological shifts while minimizing sensitivity to non-pathological factors.

Moving beyond hyperintensity thresholding. This work analyzes the challenges and outlines opportunities for advancing the field of unsupervised anomaly detection.

Implicit disentanglement of shape and appearance with federated learning for unsupervised brain pathology segmentation.