Patch2Loc

Patch2Loc introduces a learning framework for unsupervised detection and localization of brain lesions in MRI. The method learns patch-level representations and leverages predictive uncertainty to identify abnormal tissue without requiring voxel-level lesion annotations.

Oral Presentation Slides

Oral Presentation Slides

Patch2Loc Poster

Acknowledgements

Research was carried out with the support of the University of Delaware Research Foundation. This research was supported in part through the use of Information Technologies (IT) resources at the University of Delaware, specifically the high-performance computing resources. The authors would like to thank Heidi Kecskemethy and Rahul Nikam from Nemours Children’s Hospitial, Sokratis Makrogiannis from Delaware State University, and Curtis Johnson from the University of Delaware for engaging discussions regarding computer-assisted neuroradiology. We thank Finn Behrendt for providing his code as our implementation is based on and extends the publicly available code of Behrendt et al. (2025).