README.md

metadata

license: cc-by-nc-4.0
extra_gated_fields:
  Affiliation:
    type: text
    label: Your institution/affiliation
    required: true
  Intended Use:
    type: text
    label: Intended use of this dataset
    required: true
extra_gated_prompt: Please provide your affiliation and intended use to apply for access.

Model weights

Weights are published on Hugging Face: TaipingQu/BAAI-Cardiac-Agent. Extract them under the project root as weights/, consistent with expert_weight_path in app.config.

• Agent: A LLaVA multimodal model (LLaMA backbone + vision encoder), fine-tuned for cardiac MRI. Files follow the Hugging Face layout under weights/agent/.
• Expert models: Task-specific PyTorch .pth checkpoints—multi-stage segmentation for Cine 2CH / 4CH / short-axis and LGE short-axis, plus CDS (cardiac disease screening) and NICMS (non-ischemic cardiomyopathy subtyping) classifiers in per-task subfolders under weights/.

Checkpoints and layout

Subdirectory	Checkpoint	Model	Task
/	(Hugging Face layout)	LLaVA	Multimodal agent
cine_seg_first_2CH/	Cine_2CH_seg1.pth	Segmentation stage 1	Cine 2-chamber coarse
cine_seg_second_2CH/	Cine_2CH_seg2.pth	Segmentation stage 2	Cine 2-chamber refined
cine_seg_first_4CH/	Cine_4CH_seg1.pth	Segmentation stage 1	Cine 4-chamber coarse
cine_seg_second_4CH_L/	Cine_4CH_seg2_L.pth	Segmentation stage 2 (left)	Cine 4-chamber L refinement
cine_seg_second_4CH_R/	Cine_4CH_seg2_R.pth	Segmentation stage 2 (right)	Cine 4-chamber R refinement
cine_seg_first_SA/	Cine_SAX_seg1.pth	Segmentation stage 1	Cine short-axis coarse
cine_seg_second_SA/	Cine_SAX_seg2.pth	Segmentation stage 2	Cine short-axis refined
lge_seg_first_SA/	LGE_SAX_seg1.pth	Segmentation stage 1	LGE short-axis coarse
lge_seg_second_SA/	LGE_SAX_seg2.pth	Segmentation stage 2	LGE short-axis refined
diagnosis_first/	CDS.pth	Classification	Cardiac disease screening (CDS)
diagnosis_second/	NICMS.pth	Classification	NICMS subtyping

Citation

If you find our work useful in your research or application, please consider citing our paper:

@misc{qu2026baaicardiacagentintelligent,
      title={BAAI Cardiac Agent: An intelligent multimodal agent for automated reasoning and diagnosis of cardiovascular diseases from cardiac magnetic resonance imaging}, 
      author={Taiping Qu and Hongkai Zhang and Lantian Zhang and Can Zhao and Nan Zhang and Hui Wang and Zhen Zhou and Mingye Zou and Kairui Bo and Pengfei Zhao and Xingxing Jin and Zixian Su and Kun Jiang and Huan Liu and Yu Du and Maozhou Wang and Ruifang Yan and Zhongyuan Wang and Tiejun Huang and Lei Xu and Henggui Zhang},
      year={2026},
      eprint={2604.04078},
      archivePrefix={arXiv},
      primaryClass={eess.IV},
      url={https://arxiv.org/abs/2604.04078}, 
}