Abstract
Computational pathology is rapidly moving towards multimodal
and conversational systems, yet current methods communicate
diagnostic conclusions rather than localized visual evidence supporting
them, limiting explainability and evidence-grounded interaction. Pathologists
interpret tissue by identifying diagnostically relevant regions through
clinically precise descriptions that combine histological features, spatial
context, hierarchical assessment, and disambiguation from neighboring
structures. This grounded reasoning not only guides diagnosis, but also
links clinical decisions to explicit visual evidence. Yet, current methods
apply either semantic segmentation or case-level prediction, bypassing
this targeted reasoning entirely. We introduce Referral segmentation to
histopathology: a vision-language task in which models segment a tissue
region singled out by clinically grounded descriptions spanning these
four reasoning dimensions. We present ReferralSeg, a benchmark of
image–mask–instruction–quality quadruplets across breast, colon, lung,
and prostate cancers, generated through an AI-powered pipeline with
cancer-specific prompting schema and two-phase validation, readily extensible
to new organs with minimal effort. Evaluation of existing visionlanguage
segmentation models reveals substantial performance gaps, confirming
that current methods struggle to follow clinically grounded referral
instructions. ReferralSeg a foundation for more interactive, explainable
and clinically grounded computational pathology systems. The
generated dataset will be publicly released.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_071.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{AlRob_ReferralSeg_MICCAISAT2026,
author = { Al Majzoub, Roba AND Deria, Ankan AND Saeed, Numan AND Khan, Salman AND Maier-Hein, Klaus H. AND Khan, Shadab AND Khan, Fahad Shahbaz},
title = { { ReferralSeg: A Multi-Cancer Benchmark for Clinician-Style Language-Guided Segmentation in Histopathology } },
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}
}