Precise Location Matching Improves Dense Contrastive Learning in Digital Pathology

Dense prediction tasks such as segmentation and detection of pathological entities hold crucial clinical value in computational pathology workflows. However, obtaining dense annotations on large cohorts is usually tedious expensive. Contrastive learning (CL) thus often employed to leverage volumes unlabeled data pre-train the backbone network. To boost CL for prediction, some studies have proposed variations matching objectives pre-training. our analysis shows that employing existing strategies histopathology images enforces invariance among incorrect pairs features and, thus, imprecise. address this, we propose a precise location-based mechanism utilizes overlapping information between geometric transformations precisely match regions two augmentations. Extensive experiments pretraining datasets (TCGA-BRCA, NCT-CRC-HE) three downstream (GlaS, CRAG, BCSS) highlight superiority method semantic instance tasks. Our outperforms previous methods by up 7.2% average precision 5.6% Additionally, using popular contrastive frameworks, MoCo-v2, VICRegL, ConCL, improved 0.7% 5.2%, 4.0%, demonstrating generalizability. code available at https://github.com/cvlab-stonybrook/PLM_SSL .