VLM3D 2025 Task 2: Multi-Abnormality Classification

Team: Medical Image Insights

Model submitted to the VLM3D@MICCAI 2025 Multi-Abnormality Classification track.

Team Members

Yunhan Fang², Pengcheng Shi¹, Jiawei Chen³, Yiming Shi⁴, Shuheng Zhang¹, Jiaqi Liu¹, Xinglin Zhang¹

¹ Medical Image Insights · ² Shanghai Jiaotong University · ³ Xi’an Jiaotong University · ⁴ Tsinghua University

Official leaderboard

Overall rank: #3 — VLM3D 2025 CT Abnormality Classification.

https://research.forithmus.com/challenges/ct-abnormality-classification#leaderboard

Model

CT_CLS8 is a 3-checkpoint ensemble:

ct_cls8/epoch=263-crop_resume-val_F1=0.00.ckpt
ct_cls8/epoch=371-crop_cls_head-val_F1=0.00.ckpt
ct_cls8/epoch=92-crop_cls_head-f-val_F1=0.00.ckpt

Inference runs TotalSegmentator, crops seven ROIs, applies the three ROI classifiers with their class masks, and averages the sigmoid outputs.

The required TotalSegmentator checkpoint is included under:

total_segmentator/Dataset665_Totalsegmentator_v201_Anatomy/

Data

Training and evaluation use the CT-RATE dataset.

Local evaluation

CT-RATE valid_fixed, 3,039 CT volumes:

Metric Value
Macro F1 0.6129
Micro F1 0.6363
Macro precision 0.5774
Macro recall 0.6578

Local evaluation only.

Code

github.com/yinghemedical/VLM3D-2025

Report generation weights: VLM3D2025-Task1-ReportGeneration

Usage

Inference code: code/classification/inference/.

Set CT_CLS_CKPT_DIR to ct_cls8/ and point the CT_CLS8 code tree's nnUNet_results at the included TotalSegmentator directory.

Citation

If you use the CT-RATE dataset, please cite:

@article{hamamci2026generalist,
  title={Generalist foundation models from a multimodal dataset for 3D computed tomography},
  author={Hamamci, Ibrahim Ethem and Er, Sezgin and Wang, Chenyu and Almas, Furkan and Simsek, Ayse Gulnihan and Esirgun, Sevval Nil and Dogan, Irem and Durugol, Omer Faruk and Hou, Benjamin and Shit, Suprosanna and others},
  journal={Nature Biomedical Engineering},
  pages={1--19},
  year={2026},
  publisher={Nature Publishing Group UK London}
}

Data and license

CT-RATE images, reports, patient identifiers, and label CSVs are not included. Model weights are released under CC BY-NC 4.0 (non-commercial use only). TotalSegmentator and upstream dependencies retain their own licenses.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train yinghemedical/VLM3D2025-Task2-AbnormalityClassifications