Abstract
Automated report generation from 3D head computed tomography (CT) is challenging because clinical reports vary in structure and terminology, while free text provides only indirect supervision for abnormality recognition. We propose an entity-aware coarse-to-fine framework that introduces structured finding and diagnosis labels between CT understanding and report generation. Reports are first standardized into findings and assessment sections. DeepSeek-V3.2 is then used to construct a knowledge tree, extract clinical entities, and map variable expressions to normalized concepts. Each report is converted into multi-label annotations after handling polarity and uncertainty. A 3D classifier predicts these labels from CT volumes, and a large vision-language model receives both the CT and predicted labels to generate the final report. This design separates abnormality recognition from language generation and provides direct clinical supervision for 3D image understanding. Experiments on the challenge dataset demonstrate the effectiveness of the proposed framework.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HEADLINE_001.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/forum?id=c2M87QSvF9
BibTex
@InProceedings{YanSha_EntityAware_MICCAISAT2026,
author = { Yang, Shaoshuai AND Zhu, Xun AND Shi, Yiming AND Wang, Tianyi AND Chen, Xi AND Li, Miao AND Wu, Ji},
title = { { Entity-Aware Head CT Report Generation via Classification-Derived Label Priors } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}