Abstract
Automated necrosis detection in human meningioma remain underexplored in computational neuropathology, with no publicly available annotated dataset. Appearance-based representations encode tissue texture without explicitly modeling nuclear spatial distribution. This omits the histological criterion that defines necrosis and limits classification performance when labeled data is scarce. In this paper, we propose DeepNucleiNet, a nuclei-guided dual-stream framework that combines H&E tile features with binarized nuclei-map features to represent tissue morphology and nuclear spatial distribution, without additional nuclear annotations. This design directly encodes the histological criterion that defines necrosis, grounding the model representation in tissue biology rather than appearance-based texture patterns. DeepNucleiNet was evaluated under strict inter-patient cross-validation across eight meningioma patients, followed by external validation without retraining on TCGA GBM, TCGA LGG, DeepHisto, and TiGER. DeepNucleiNet achieved a mean F1-score of 0.942 across four inter-patient folds, higher than the pathology foundation models UNI2-h and Virchow2 (0.884 and 0.730, respectively) and the appearance-only XceptionNet baseline (0.847). Without retraining, the meningioma-trained model achieved F1-scores of 0.943, 0.843, and 0.635 on TCGA GBM, DeepHisto, and TiGER, respectively, with no necrosis predictions on TCGA LGG slides. These results indicate that directly encoding nuclear depletion improves necrosis detection performance and generalizes across related CNS tumor types.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BrainWorks_033.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/profile?id=~Dasari_Naga_Raju1
BibTex
@InProceedings{RajDas_DeepNucleiNet_MICCAISAT2026,
author = { Raju, Dasari Naga AND Srikanth, T. K. AND Rao, Shilpa AND Kestur, Ramesh AND A., Mahadevan},
title = { { DeepNucleiNet: Encoding Spatial Nuclear Patterns for Data-Efficient Necrosis Detection in Computational Neuropathology } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17254},
month = {pending},
page = {pending}
}