Abstract
Cell detection and classi cation are fundamental tasks in
computational pathology. Recent DETR-based methods formulate nucleus
analysis as direct object detection, simplifying the prediction pipeline
while achieving competitive performance. Although larger elds of view
improve cell classi cation by providing broader tissue context, scaling
a single DETR detector requires substantially more object queries, increasing
computational cost.
We propose Dual-CellNucDETR, a dual-stage Deformable DETR architecture
that decouples nucleus localisation from contextual classi cation.
A rst stage detects nuclei on local image patches, while a second stage
performs contextual classi cation over all detected nuclei using the complete
image, enabling larger elds of view without scaling the detector.
Experiments on PUMA and Internal Lung histopathology datasets show
that the proposed approach achieves competitive or superior classi -
cation performance while substantially reducing memory consumption
compared with a single-stage DETR operating on full-resolution images.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_052.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{YseMar_DualCellNucDETR_MICCAISAT2026,
author = { Ysern, Maria AND Vila-Bagaria, Sigrid AND Pina, Oscar AND Vilaplana, Verónica},
title = { { Dual-CellNucDETR: Context-Aware Object-Level Cell Analysis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17251},
month = {pending},
page = {pending}
}