You need to agree to share your contact information to access this model
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
These weights are released for non-commercial research use. Please describe your intended use before requesting access; requests are reviewed manually.
Log in or Sign Up to review the conditions and access this model content.
LungEvaty-CL-MIL — LIDC Malignancy Classification Weights
This repository hosts the pre-trained weights for the LungEvaty-CL-MIL classifier: a scan-level malignancy classifier for lung CT built on top of the LungEvaty transformer backbone with a lightweight gated-attention multiple-instance learning (MIL) head trained on LIDC-IDRI.
- Project page: https://maulikchevli.github.io/lungevaty/
- Code: https://github.com/jawbra/lungevaty_cl_mil
About LungEvaty-CL-MIL
LungEvaty-CL-MIL predicts a single scan-level malignancy probability from a lung CT. The frozen encoder is the LungEvaty transformer pretrained on NLST for 1–6-year risk prediction; on top of it we train a gated-attention MIL head that aggregates per-nodule instance features into a scan-level score.
The MIL head is small (~800 KB) and expects LungEvaty features as input, so the same backbone weights can be re-used across the survival and classification tasks. This checkpoint reports AUC ≈ 0.83 (AP ≈ 0.44) on the internal LIDC test split (157 scans, 23 positive / 134 negative) and was the best-mean configuration across seeds in an ablation over encoder-unfreezing depth.
Files
| File | Description |
|---|---|
lungevaty_cl_mil_best.pt |
MIL-head weights (frozen LungEvaty backbone), best-seed checkpoint. |
Usage
Download the weights and use them with the code repo:
huggingface-cli download jawbra/lungevaty-cl-mil lungevaty_cl_mil_best.pt \
--local-dir ./weights
Then follow the instructions in the code repository for loading the LungEvaty backbone and attaching this MIL head.
Access
This repository is public but gated — each download request is reviewed manually. Please state your affiliation and intended use when requesting access.
Citation
If you use these weights, please cite:
@inproceedings{brandt2026lungevaty,
title={LungEvaty: A scalable, open-source transformer-based deep learning model for lung cancer risk prediction in LDCT screening},
author={Brandt, Johannes and Chevli, Maulik and Braren, Rickmer and Kaissis, Georgios and M{\"u}ller, Philip and Rueckert, Daniel},
booktitle={2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI)},
pages={1--5},
year={2026},
organization={IEEE}
}