CBCT-guided Adaptive Radiotherapy using Self-Supervised Sequential Domain
Description
Adaptive radiotherapy (ART) is an advanced technology in modern cancer treatment that incorporates progressive changes in patient anatomy into active plan/dose adaption during the fractionated treatment. However, the clinical application relies on the accurate segmentation of cancer tumors on low-quality on-board images, which has posed challenges for both manually delineation and deep learning-based models. In this paper, we propose a novel sequence transduction deep neural networks with attention mechanism to learn the shrinkage of the cancer tumor based on patients' weekly cone-beam computed tomography (CBCT). We design a self-supervised domain adaption (SDA) method to learn and adapt the rich textural and spatial features from pre-treatment high-quality computed tomography (CT) to CBCT modality in order to address the poor image quality and lack of labels. We also provide uncertainty estimation for sequential segmentation, which aids not only in the risk management of treatment planning but also in the calibration and reliability of the model. Our experimental results based on a clinical non-small cell lung cancer (NSCLC) dataset with six-teen patients and ninety-six longitudinal CBCTs show that our model properly learns weekly deformation of the tumor over time with an average dice score of 0.92 on the immediate next step, and is able to predict multiple steps (up to 5 weeks) for future patient treatments with an average dice score reduction of 0.05. By incorporating the tumor shrinkage predictions into a weekly re-planning strategy, our proposed method demonstrates a significant reduction in the risk of radiation-induced pneumonitis up to 35% while maintaining the high tumor control probability. The code that supports the findings of the study is openly available in the repository, patient data is available on request due to institutional restrictions.
Files
Steps to reproduce
Requirements: Hydra==2.5 hydra-core==1.1.1 keras==2.6.0 Keras-Applications==1.0.8 keras-nightly==2.5.0.dev2021032900 Keras-Preprocessing==1.1.2 MedPy==0.4.0 notebook==6.2.0 numpy==1.19.5 nvidia-dali-cuda110==1.3.0 nvidia-dali-tf-plugin-cuda110==1.3.0 nvidia-dlprofviewer @ file:///opt/dlprof_viewer_install/nvidia_dlprofviewer-1.3.0-py3-none-any.whl nvtx-plugins==0.1.8 omegaconf==2.1.1 pandas==1.3.2 PyYAML==5.4.1 scikit-image==0.18.3 scikit-learn==0.24.2 scipy==1.4.1 sklearn==0.0 tensorboard-data-server==0.6.1 tensorboard-plugin-wit==1.8.0 tensorflow==2.6.0 tensorflow-addons @ file:///opt/tensorflow/tf-addons/artifacts/tensorflow_addons-0.13.1-cp38-cp38-linux_x86_64.whl tensorflow-datasets==3.2.1 tensorflow-estimator==2.6.0 tensorflow-examples===af9e301238f7f5e939c4adea52eaee7036d28e95- tensorflow-hub==0.12.0 tensorflow-metadata==1.1.0 tensorflow-model-optimization==0.6.0 tensorflow-text==2.6.0 tf-models-official==2.6.0 tf-slim==1.1.0 torch==1.9.0 run data_util.py to preprocess all datasets -> saves these files: data_list, ct, cbct, and patientweek2xyz files with the following format: {conf.name}_{conf.patch_size_im}_{conf.patch_size_sg}_{conf.xyzmargin} modules inlcude models and customized layers for spatio-temporal segmentation as well as the uncertainty estimation. Configurations are used to hyper-tune the model and also to compute the uncertainty of the ensemble models.