MMedAgent/mmedagent-plus-demo-cases

MMedAgent-Plus demo cases

The 25 opening requests of the public MMedAgent-Plus demo (paper, arXiv:2610.27233): the question a user sends in the first turn and the image(s) sent with it. 22 cases carry an image (20 distinct files), 3 are text-only. Together they exercise all 18 specialist tools of the agent across radiology (X-ray, CT, MRI, ultrasound), pathology, ophthalmology and text.

Each case is one row. kind is single-tool for the 19 single-tool examples of the demo's example library and story-opener for the first turn of its 6 multi-turn stories (the follow-up turns, the tool outputs and the agent's answers are not part of this dataset).

Fields

Field	Meaning
case_id	id used by the demo (skills:N, multi:N)
kind	single-tool or story-opener
title	the case's title in the demo (card label, or the story title)
tools	the specialist tool(s) the request is meant to exercise
question	the user's first message, verbatim as sent to the agent
images	the image(s) sent with the message: none for the text-only tools, one for most cases, two for the registration case
image_sources	one entry per image, same order: source dataset, source file name, how the bundled file relates to it, licence, URL, sha256
source_split_file, source_row, source_turn	the row of the MMedAgent-Plus instruction-tuning data the question comes from (0-based row in that JSONL file)
question_verbatim	whether the text is that row's wording unchanged
origin	how the case was derived from the row (verbatim, rephrased question, replacement image, …)

Cases

case_id	kind	tool(s)	image	question
skills:0	single-tool	BiomedCLIP	biomedclip_chest_xray.jpg	What type of imaging modality is used to acquire the above image?
skills:1	single-tool	MedSAM	medsam_kidney_flare7.jpg	Segment the kidney inside the box [0.56, 0.55, 0.78, 0.72].
skills:2	single-tool	Grounding DINO	gdino_1.jpg	Can you identify the lungs in this 2D x-ray chest scan and provide their locations?
skills:3	single-tool	Grounding DINO + MedSAM	gdino_medsam_1.jpg	Can you identify and segment the tumor in this brain MRI image?
skills:4	single-tool	UltraSAM	ultrasam_1.png	Segment the lesion inside the box [0.39, 0.11, 0.69, 0.41] in this ultrasound image.
skills:5	single-tool	IterNet	iternet_papila5.png	Please identify the vessel in this retina image.
skills:6	single-tool	UniGradICON	unigradicon_flare7_43.jpg, unigradicon_flare7_55.jpg	Help me align these images for analysis.
skills:7	single-tool	HealthGPT	healthgpt_brats.jpg	Produce a more detailed image.
skills:8	single-tool	LLaVA-Med	llavamed_1.jpg	What organ is shown in the image?
skills:9	single-tool	LLaVA-Rad	llavarad_mcucxr_0126.jpg	Write a complete radiology report including impressions and recommendations.
skills:10	single-tool	SpecialistVLMs	specialistvlms_1.jpg	Summarize this OCT image into a clinical report.
skills:11	single-tool	CONCH	conch_norm_mucosa.jpg	What is present in the image?
skills:12	single-tool	DSMIL-C16	dsmilc16_1.jpg	Determine whether this image is malignant or benign.
skills:13	single-tool	DSMIL-TCGA	dsmiltcga_1.jpg	Please classify this pathology image of lung tissue.
skills:14	single-tool	CellViT	cellvit_1.jpg	How many cells are present in this pathology image? Image format is svs.
skills:15	single-tool	CellSAM	cellsam_pcam.jpg	Segment the cells in the tissue sample and give me the labeled mask.
skills:16	single-tool	PMC-LLaMA	(text only)	I'm giving you evidence for a biomedical query. How would you answer it? ### Question: Does anacetrapib lo…
skills:17	single-tool	RaTE-NER	(text only)	Identify and label medical entities including anatomical and pathological terms. ### Text to analyze: trac…
skills:18	single-tool	ChatCAD-R	(text only)	I've just been diagnosed with Diabetic Retinopathy. Could you tell me more about it and the steps I need to…
multi:0	story-opener	BiomedCLIP, Grounding DINO + MedSAM, LLaVA-Rad, RaTE-NER	cxr_mcucxr_0203.jpg	What type of imaging modality is used to acquire this image?
multi:1	story-opener	BiomedCLIP, Grounding DINO, MedSAM, RaTE-NER	medsam_kidney_flare7.jpg	What type of imaging modality is used to acquire this image?
multi:2	story-opener	BiomedCLIP, Grounding DINO + MedSAM, RaTE-NER	gdino_medsam_1.jpg	What imaging modality is this?
multi:3	story-opener	CONCH, CellViT, PMC-LLaMA	tum_adenocarcinoma.jpg	What is present in the image?
multi:4	story-opener	BiomedCLIP, IterNet, ChatCAD-R	iternet_papila.png	What kind of medical image is this?
multi:5	story-opener	BiomedCLIP, UltraSAM, PMC-LLaMA	ultrasam_1.png	What imaging modality is this?

How the cases were made

Questions are taken from the MMedAgent-Plus test split (a few from the training split); where question_verbatim is false the wording was lightly rephrased, for example to pass a normalized bounding box. Images are the dataset's own image for that row whenever its licence allows redistribution; otherwise an image from a redistributable source of the same modality replaces it (origin says so, and image_sources.relation_to_source says whether the bundled file is byte-identical, resized or converted). Files are served unchanged from the demo (https://mmedagent.org/demo/).

Image sources and licences

The text of this dataset is released under CC BY 4.0. Every image keeps the licence of its source dataset, listed below; cite the source when you reuse an image. One file (specialistvlms_1.jpg) is CC BY-NC-ND 4.0, i.e. non-commercial use only and no modification.

Dataset	Licence	Files	Citation
Montgomery County chest X-ray set (NLM, MCUCXR)	Public dataset released by the U.S. National Library of Medicine; citation requested	llavarad_mcucxr_0126.jpg, cxr_mcucxr_0203.jpg	Jaeger S, Candemir S, Antani S, Wáng YX, Lu PX, Thoma G. Two public chest X-ray datasets for computer-aided screening of pulmonary diseases. Quant Imaging Med Surg 2014;4(6):475-477.
Shenzhen Hospital chest X-ray set (NLM, CHNCXR)	Public dataset released by the U.S. National Library of Medicine; citation requested	gdino_1.jpg	Jaeger S, et al. Two public chest X-ray datasets for computer-aided screening of pulmonary diseases. Quant Imaging Med Surg 2014;4(6):475-477.
PAPILA fundus dataset	CC BY 4.0	iternet_papila.png, iternet_papila5.png	Kovalyk O, et al. PAPILA: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment. Sci Data 2022;9:291.
NCT-CRC-HE-100K / CRC-VAL-HE-7K histology patches	CC BY 4.0	cellvit_1.jpg, conch_norm_mucosa.jpg, tum_adenocarcinoma.jpg	Kather JN, Halama N, Marx A. 100,000 histological images of human colorectal cancer and healthy tissue. Zenodo 2018.
PatchCamelyon (PCam)	CC0 1.0	dsmilc16_1.jpg, dsmiltcga_1.jpg, cellsam_pcam.jpg	Veeling BS, Linmans J, Winkens J, Cohen T, Welling M. Rotation equivariant CNNs for digital pathology. MICCAI 2018. Derived from Camelyon16 (CC0).
BraTS brain MRI	CC BY 4.0 (BraTS 2021 release)	gdino_medsam_1.jpg, healthgpt_brats.jpg	Baid U, et al. The RSNA-ASNR-MICCAI BraTS 2021 benchmark on brain tumor segmentation and radiogenomic classification. arXiv:2107.02314.
FLARE22 abdominal CT (labelled training cases)	CC BY 4.0 (Zenodo release of the 50 labelled training cases; images originate from the Medical Segmentation Decathlon, CC BY-SA 4.0)	medsam_kidney_flare7.jpg, unigradicon_flare7_43.jpg, unigradicon_flare7_55.jpg	Ma J, et al. Unleashing the strengths of unlabeled data in pan-cancer abdominal organ quantification: the FLARE22 challenge. Lancet Digit Health 2024.
BrEaST breast ultrasound (TCIA Breast-Lesions-USG)	CC BY 4.0	ultrasam_1.png	Pawłowska A, et al. Curated benchmark dataset for ultrasound based breast lesion analysis. Sci Data 2024;11:148.
OCT5k retinal OCT	CC BY-NC-ND 4.0 (non-commercial use only; image unmodified)	specialistvlms_1.jpg	Arikan M, et al. OCT5k: a dataset of multi-disease and multi-graded annotations for retinal layers. Sci Data 2025;12:267.
PubMed Central open-access article figures	CC BY (per article)	biomedclip_chest_xray.jpg, llavamed_1.jpg	biomedclip_chest_xray.jpg: Lung India 2010;27(3), PMC2946729, Fig. 1 (CC BY). llavamed_1.jpg: Cureus 2021;13(10):e19055, PMC8614211, Fig. 4 (CC BY).

Citation

@misc{mmedagentplus2026,
  title   = {MMedAgent-Plus: Scaling Multimodal Medical Agents
             across Modalities, Tasks and Workflows},
  author  = {Ruoxin Wang and Yuanting Pan and Binxu Li and Ruiyang Ji
             and Yaqian Chen and Jiayuan Ding and Qifan Wang and Lifu
             Huang and Mingjie Li and Yu Zhang and Ehsan Adeli and
             Kilian Pohl and Zhe Xu and Zihao Lin and Haoyu Dong and
             Yixin Wang},
  year    = {2026},
  eprint  = {2610.27233},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url     = {https://arxiv.org/abs/2610.27233}
}