Seeing Beyond the Lesion: Disease Recognition from Reactive CNS Tissue
Abstract
Sampling error yields exclusively reactive, non-lesional brain parenchyma in a significant proportion of intracranial biopsies, leaving the underlying disease undiagnosed. We benchmark four pathology foundation models (UNI2-h, Virchow2, Prov-GigaPath, H-optimus-0) as frozen patch encoders within a shared attention-based multiple-instance learning framework using 245 whole-slide images from 186 patients with confirmed downstream diagnoses. We first show that coarse disease-category prediction can be reproduced largely from slide size alone. After restricting classification to three finer diagnostic distinctions within common tissue categories, this confound no longer explains performance, yet disease labels remain predictable above chance under permutation testing (p ≤ 10−4 throughout). Surprisingly, performance is statistically indistinguishable across all foundation-model encoders, suggesting that recovering these weak morphological signatures is not limited by current patch representations. Signed instance-contribution maps and expert review further test whether predictive evidence localizes to reactive parenchyma rather than sampling-induced bias like blood introduced during tissue sampling. These results position acquisition-shortcut auditing via a provenance-only baseline as a necessary control in computational-pathology benchmarks, and show, once that confound is removed, that weakly supervised models still recover disease signal from tissue conventionally regarded as non-diagnostic.
Related
- Seeing the invisible: Reorienting cardiovascular risk prediction across the disease trajectory
- 1141 Spatial immune profiling across the lobular breast cancer continuum reveals progressive immune exclusion from precursor lesion to invasive disease
- Beyond one-to-one mappings: Modelling distributed lesion-symptom relationships with multilayer networks
- Beyond Darwin: reactive heredity, burst-drift dynamics and eco‑evolutionary control of cancer
- Beyond Initial Crisis Recognition: Suicide-Risk Responses of ChatGPT, Claude, and Gemini Across 14-Turn Ukrainian-Language Conversations
BibTeX
@misc{schnorrenberg2026seeing,
title = {{Seeing Beyond the Lesion: Disease Recognition from Reactive CNS Tissue}},
author = {Jan Schnorrenberg and Jan Ernsting and Enrico Küllenberg and Tim Hahn and Benjamin Risse and Christian Thomas},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.02390},
eprint = {2609.02390},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.02390},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Seeing Beyond the Lesion: Disease Recognition from Reactive CNS Tissue
AU - Jan Schnorrenberg
AU - Jan Ernsting
AU - Enrico Küllenberg
AU - Tim Hahn
AU - Benjamin Risse
AU - Christian Thomas
PY - 2026
DA - 2026/09/02
PB - arXiv
DO - 10.48550/arxiv.2609.02390
UR - https://arxiv.org/abs/2609.02390
AB - Sampling error yields exclusively reactive, non-lesional brain parenchyma in a significant proportion of intracranial biopsies, leaving the underlying disease undiagnosed. We benchmark four pathology foundation models (UNI2-h, Virchow2, Prov-GigaPath, H-optimus-0) as frozen patch encoders within a shared attention-based multiple-instance learning framework using 245 whole-slide images from 186 patients with confirmed downstream diagnoses. We first show that coarse disease-category prediction can be reproduced largely from slide size alone. After restricting classification to three finer diagnostic distinctions within common tissue categories, this confound no longer explains performance, yet disease labels remain predictable above chance under permutation testing (p ≤ 10^−4 throughout). Surprisingly, performance is statistically indistinguishable across all foundation-model encoders, suggesting that recovering these weak morphological signatures is not limited by current patch representations. Signed instance-contribution maps and expert review further test whether predictive evidence localizes to reactive parenchyma rather than sampling-induced bias like blood introduced during tissue sampling. These results position acquisition-shortcut auditing via a provenance-only baseline as a necessary control in computational-pathology benchmarks, and show, once that confound is removed, that weakly supervised models still recover disease signal from tissue conventionally regarded as non-diagnostic.
ER -