Patch2Loc: Learning to Localize Patches for Unsupervised Brain Lesion Detection

2025-06-25

AI Summary

This work addresses unsupervised anomaly detection in brain MRI, where no abnormal samples or pixel-level annotations are available. We propose Patch2Loc, a self-supervised framework that models spatial positional relationships among normal image patches to localize anomalies via prediction error and its variance—yielding pixel-level anomaly heatmaps for precise lesion localization and segmentation (e.g., tumors, white matter hyperintensities). To our knowledge, this is the first method to formulate patch localization as a core task in unsupervised medical image anomaly detection. By fusing error-based heatmaps with pixel-wise reconstruction outputs, Patch2Loc significantly improves segmentation accuracy. Extensive evaluation on four public benchmarks—BraTS2021, MSLUB, ATLAS, and WMH—demonstrates consistent superiority over existing unsupervised approaches, establishing new state-of-the-art performance across all datasets.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Segmentation
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Detecting brain lesions as abnormalities observed in magnetic resonance imaging (MRI) is essential for diagnosis and treatment. In the search of abnormalities, such as tumors and malformations, radiologists may benefit from computer-aided diagnostics that use computer vision systems trained with machine learning to segment normal tissue from abnormal brain tissue. While supervised learning methods require annotated lesions, we propose a new unsupervised approach (Patch2Loc) that learns from normal patches taken from structural MRI. We train a neural network model to map a patch back to its spatial location within a slice of the brain volume. During inference, abnormal patches are detected by the relatively higher error and/or variance of the location prediction. This generates a heatmap that can be integrated into pixel-wise methods to achieve finer-grained segmentation. We demonstrate the ability of our model to segment abnormal brain tissues by applying our approach to the detection of tumor tissues in MRI on T2-weighted images from BraTS2021 and MSLUB datasets and T1-weighted images from ATLAS and WMH datasets. We show that it outperforms the state-of-the art in unsupervised segmentation. The codebase for this work can be found on our href{https://github.com/bakerhassan/Patch2Loc}{GitHub page}.

Problem

• Localizing brain lesions without annotated training data
• Detecting abnormal patches via spatial prediction errors
• Improving unsupervised segmentation of brain MRI abnormalities

Innovation

• Unsupervised learning from normal MRI patches
• Neural network mapping patches to spatial locations
• Heatmap generation for abnormal patch detection