Abstract

Contrast-enhanced computed tomography (CECT) is a pivotal tool for tumor diagnosis and longitudinal disease assessment. During follow-up evaluations, radiologists must compare current and prior scans to summarize tumor progression—a labor-intensive and cognitively demanding task. We introduce the first vision-language framework for automated abdominal tumor follow-up report generation from longitudinal CT imaging. To facilitate this research, we curate a large-scale clinical dataset comprising thousands of cancer patients with multi-phase CT examinations and structured follow-up reports detailing disease evolution. We propose a novel 3D vision-language model that jointly ingests the previous CT volume, current CT volume, and prior radiology report to generate a follow-up report reflecting interval tumor changes. Unlike traditional 2D slice-based approaches, our method utilizes a 3D vision encoder to process volumetric CT data directly, ensuring spatial understanding of the anatomy. To mitigate coarse-grained reasoning and clinical hallucinations, we implement an organ-level fine-grained alignment strategy by training auxiliary organ-specific follow-up generation tasks grounded in automatically segmented liver, kidney, pancreas, and lung regions.Furthermore, we integrate a tumor-aware regression objective that predicts relative tumor burden within each organ, providing quantitative supervision for disease progression. Experiments demonstrate that our approach significantly outperforms both 2D and 3D vision-language baselines across multiple automatic evaluation metrics, providing compelling evidence for the effectiveness of fine-grained clinical supervision.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_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/forum?id=EQUWZx11Wx

BibTex

@InProceedings{CheYan_OrganAware_MICCAISAT2026,
        author = { Chen, Yan AND Ouyang, Xi AND Xue, Zhong AND Liu, Aie AND Wu, Guangyu AND Zhu, Mengying AND Shi, Feng AND Gu, Dongdong AND Shen, Dinggang AND Wang, Qian},
        title = { { Organ-Aware Longitudinal Report Generation for Abdominal Tumor Assessment from 3D Contrast-Enhanced CT } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17272},
        month = {pending},
        page = {pending}
}


back to top