File size: 2,809 Bytes
d479880
 
ebba391
 
 
 
 
 
 
 
 
d479880
ebba391
 
 
 
 
 
 
 
 
 
 
 
00a14c3
8284cf2
ebba391
 
 
 
 
 
 
 
 
 
 
 
 
 
bf90f51
ebba391
bf90f51
ebba391
 
 
 
 
 
 
8284cf2
ebba391
8284cf2
 
 
 
 
ebba391
 
 
 
 
8284cf2
bf90f51
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
---
license: cc-by-nc-4.0
pipeline_tag: image-segmentation
tags:
  - medical-imaging
  - ct
  - interstitial-lung-disease
  - airway-segmentation
  - nnunet
  - medsam2
  - monai-label
---

# ILD-SemiSegTool: model weights

Model weights for the paper **"Prompt-Guided Interactive Segmentation of Interstitial Lung Disease in Thoracic CT"**, accepted at TIA 2026 @ MICCAI 2026.

- Paper: https://arxiv.org/abs/2608.28453
- Code: https://github.com/AIHNlab/ILD-SemiSegTool

## Models

| Model | Task | Files |
|---|---|---|
| nnU-Net (Lung, ILD, ILD raw(without masking the lung from CT)) | Automatic Lung, ILD segmentation in thoracic CT | `[(https://huggingface.co/VasilisDedousis/ILD-SemiSegTool/tree/main/HuggingFace/nnUNet_weights)]` |
| MedSAM2-ILD | Prompt-guided interactive ILD refinement | `[(https://huggingface.co/VasilisDedousis/ILD-SemiSegTool/tree/main/HuggingFace/MedSAM2%2BnnUNet_weights)]` |

The models are served through a MONAI Label radiology app with an OHIF viewer and Orthanc storage. See the GitHub repository for setup.

## Usage

```python
from huggingface_hub import snapshot_download
snapshot_download("VasilisDedousis/ILD-SemiSegTool", local_dir="weights")
```

Then follow the setup instructions in the GitHub README to place the weights in the MONAI Label app.

## Training data

## Training data

The models were trained on 306 thoracic CT scans (1 mm slices, hard-kernel reconstruction) from patients with fibrotic and obstructive lung disease, drawn from the LTRC repository and the cohort of Christe et al. (2019). Annotations cover seven ILD patterns (GGO, reticulation, consolidation, honeycombing, reticulation+GGO, traction bronchiectasis, emphysema) plus healthy tissue. Data were split 80/10/10 at scan level. See the [paper](https://arxiv.org/abs/2608.28453) for details.
## Intended use

For research use only. These models are not validated for clinical use and must not be used for diagnosis or treatment decisions.

## Citation

```bibtex
@inproceedings{dedousis2026promptguided,
  title     = {Prompt-Guided Interactive Segmentation of Interstitial Lung Disease in Thoracic CT},
  author    = {Dedousis, Vasilis and Abdur Rahman, Lubnaa and Brigato, Lorenzo and Dack, Ethan and Christe, Andreas and Frank, Christoph and Funke-Chambour, Manuela and Roos, Justus and Huber, Adrian and Ebner, Lukas and Mougiakakou, Stavroula},
  booktitle = {Thoracic Image Analysis (TIA) Workshop, MICCAI 2026},
  year      = {2026},
  url       = {https://papers.miccai.org/miccai-2026-sat/TIA_022.html},
  note      = {arXiv:2608.28453}
}
```

## Contact

AI in Health and Nutrition (AIHN) Lab, ARTORG Center, University of Bern.
https://www.artorg.unibe.ch/research/aihn/index_eng.html

For questions, please open an [issue](https://github.com/AIHNlab/ILD-SemiSegTool/issues).