From Linear Probing to Joint-Weighted Token Hierarchy: A Foundation Model Bridging Global and Cellular Representations in Biomarker Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Jingsong, Li, Han, Navab, Nassir, Schüffler, Peter J.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911253699493888
author Liu, Jingsong
Li, Han
Navab, Nassir
Schüffler, Peter J.
author_facet Liu, Jingsong
Li, Han
Navab, Nassir
Schüffler, Peter J.
contents AI-based biomarkers can infer molecular features directly from hematoxylin & eosin (H&E) slides, yet most pathology foundation models (PFMs) rely on global patch-level embeddings and overlook cell-level morphology. We present a PFM model, JWTH (Joint-Weighted Token Hierarchy), which integrates large-scale self-supervised pretraining with cell-centric post-tuning and attention pooling to fuse local and global tokens. Across four tasks involving four biomarkers and eight cohorts, JWTH achieves up to 8.3% higher balanced accuracy and 1.2% average improvement over prior PFMs, advancing interpretable and robust AI-based biomarker detection in digital pathology.
format Preprint
id arxiv_https___arxiv_org_abs_2511_05150
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Linear Probing to Joint-Weighted Token Hierarchy: A Foundation Model Bridging Global and Cellular Representations in Biomarker Detection
Liu, Jingsong
Li, Han
Navab, Nassir
Schüffler, Peter J.
Computer Vision and Pattern Recognition
Artificial Intelligence
AI-based biomarkers can infer molecular features directly from hematoxylin & eosin (H&E) slides, yet most pathology foundation models (PFMs) rely on global patch-level embeddings and overlook cell-level morphology. We present a PFM model, JWTH (Joint-Weighted Token Hierarchy), which integrates large-scale self-supervised pretraining with cell-centric post-tuning and attention pooling to fuse local and global tokens. Across four tasks involving four biomarkers and eight cohorts, JWTH achieves up to 8.3% higher balanced accuracy and 1.2% average improvement over prior PFMs, advancing interpretable and robust AI-based biomarker detection in digital pathology.
title From Linear Probing to Joint-Weighted Token Hierarchy: A Foundation Model Bridging Global and Cellular Representations in Biomarker Detection
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2511.05150