Multi-model Study of Fast VMAT Segment Dose Calculation with Deep Learning

This repository contains the source code, model weights, and Dockerfile accompanying the paper:

Multi-model study of fast VMAT segment dose calculation with deep learning Fan Xiao, Niklas Wahl, Claus Belka, Christopher Kurz, Georgios Dedes, Guillaume Landry Physics in Medicine & Biology, 2026. https://doi.org/10.1088/1361-6560/ae6413

Repository Contents

• docker/: Dockerfile for setting up a reproducible environment.
• preprocessing/: Scripts for BEV cuboid/patient coordinate (four physical inputs) processing.
• model/: Source code and weights for the CNN-ConvLSTM, CNN-Mamba, DoTA(pytorch), C3D, DeepDose-C3D.
• train/: Shared training utils (utils.py) and per-model entry points (CNN_ConvLSTM/train.py, CNN_Mamba/train.py).
• inference/: Shared inference pipeline (pipeline.py) and per-model entry points (CNN_ConvLSTM/inference.py, CNN_Mamba/inference.py).

Model Architectures

Inference

# CNN-ConvLSTM
python inference/CNN_ConvLSTM/inference.py \
    --patient PATIENT_ID \
    --sim-root /path/to/simulation \
    --seg-dir /path/to/segments \
    --ct-root /path/to/ct \
    --model-weights /path/to/weights.pth

# CNN-Mamba
python inference/CNN_Mamba/inference.py \
    --patient PATIENT_ID \
    --sim-root /path/to/simulation \
    --seg-dir /path/to/segments \
    --ct-root /path/to/ct \
    --model-weights /path/to/weights.pth

Citation

If you find this code useful in your research, please consider citing:

@article{xiao2026multimodel,
  title   = {Multi-model study of fast {VMAT} segment dose calculation with deep learning},
  author  = {Xiao, Fan and Wahl, Niklas and Belka, Claus and Kurz, Christopher and Dedes, Georgios and Landry, Guillaume},
  journal = {Physics in Medicine \& Biology},
  year    = {2026},
  doi     = {10.1088/1361-6560/ae6413}
}

Open Challenge

An associated open challenge is available at the DoseRad 2026 Grand Challenge.