LungEvaty — Official Model Weights
This repository hosts the official pre-trained weights for LungEvaty, a fully Transformer-based framework for 1–6 year lung cancer risk prediction from a single low-dose CT (LDCT) scan.
- Project page: https://maulikchevli.github.io/lungevaty/
- Code: https://github.com/maulikchevli/lungevaty
About LungEvaty
LungEvaty predicts 1–6 year lung cancer risk from whole-lung LDCT inputs, using a fully Transformer-based architecture that learns directly from large-scale screening data without requiring pixel-level region supervision. It matches state-of-the-art performance using imaging data alone, and can be further refined with an optional Anatomically Informed Attention Guidance (AIAG) loss that encourages anatomically focused attention.
The model was trained on more than 90,000 CT scans, with over 28,000 used for fine-tuning and 6,000 for evaluation. The framework is designed as a simple, data-efficient, and fully open-source foundation for future research in longitudinal and multimodal lung cancer risk prediction.
Files
| File | Description |
|---|---|
lungevaty_best.pt |
Best survival checkpoint for inference with dcm_inference.py. |
Usage
Clone the code repository, then point the inference script at the downloaded checkpoint:
huggingface-cli download jawbra/lungevaty lungevaty_best.pt --local-dir ./weights
python dcm_inference.py \
log.ckpt_loc=./weights \
testing.use_checkpoint=lungevaty_best.pt
See dcm_inference.py and the sample manifest in the code repo for the expected input schema.
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}
}