Invertible and Variable Augmented Network for Pretreatment Patient-Specific Quality Assurance Dose Prediction
Zhongsheng Zou1, Changfei Gong2, Lingpeng Zeng1
1Department of Electronic Information Engineering, Nanchang University, Nanchang, China.
Journal of Imaging Informatics in Medicine
|February 12, 2024
Summary
A novel deep learning (DL) network, IVPSQA, improves radiotherapy quality assurance efficiency. This AI tool accurately predicts dose distributions, reducing medical physicists' workload.
Area of Science:
- Medical Physics
- Radiotherapy
- Artificial Intelligence
Background:
- Pretreatment patient-specific quality assurance (prePSQA) is crucial for verifying radiotherapy accuracy.
- Current prePSQA measurement methods are time-consuming and increase medical physicist workload.
Purpose of the Study:
- To develop a novel deep learning (DL) network to enhance the accuracy and efficiency of prePSQA.
- Introduce the Invertible and Variable augmented network for prePSQA (IVPSQA).
Main Methods:
- A modified invertible and variable augmented network (IVPSQA) was developed.
- Trained on 240 volumetric modulated arc therapy (VMAT) patient cases (CT images, planned dose, measured dose) and tested on 60.
- Utilized Adam algorithm for optimization.
Main Results:
- IVPSQA accurately predicted 3D prePSQA dose distributions for head and neck, chest, and abdomen cancers.
- Outperformed U-Net-based methods in dose difference maps and horizontal profiles.
- Achieved high agreement with ground truth (SSIM, MSE, MAE).
Conclusions:
- The IVPSQA model offers a promising approach for predicting 3D prePSQA dose distributions.
- This AI tool can serve as a clinical guidance system, significantly reducing prePSQA measurement efforts for medical physicists.
Keywords:
Deep learningInvertible and variable augmented networkPretreatment patient-specific quality assuranceRadiotherapyVolumetric modulated arc therapy
