CPFTransGAN: A Cross Perception Fusion Transformer-Based Generative Adversarial Network for Head and Neck Cancer Dose...

Area of Science:

• Medical Physics
• Artificial Intelligence in Medicine
• Oncology

Background:

• Radiation therapy is crucial for head and neck cancer, requiring precise dose delivery to tumors while sparing healthy tissues.
• Accurate quantitative dose prediction is essential for advancing precision radiotherapy.
• Current methods face challenges in optimizing dose distribution and organ sparing.

Purpose of the Study:

• To develop an advanced generative adversarial network for improved radiation dose prediction in head and neck cancer.
• To enhance the accuracy of predicting radiation dose distributions for planning target volumes (PTV) and organs at risk (OAR).
• To integrate deep learning techniques for more intelligent and precise radiotherapy planning.

Main Methods:

• Proposed a novel generative adversarial network, CPFTrans-GAN, utilizing a Cross Perception Fusion Transformer (CPF Transformer) module.
• Developed a CPF Transformer-based generator (CPFTransGenerator) with a four-stage encoding-decoding structure.
• Implemented an adaptive weight loss for discriminator training and a multiscale cross-window encoding network for enhanced prediction accuracy.

Main Results:

• CPFTrans-GAN demonstrated superior performance in quantitative dose prediction compared to state-of-the-art methods.
• The method achieved high accuracy in predicting dose distributions on public and clinical head and neck cancer datasets.
• The integrated CPF Transformer module effectively improved the fusion of CNN and Transformer capabilities.

Conclusions:

• The proposed CPFTrans-GAN offers a significant advancement in radiation dose prediction for head and neck cancer radiotherapy.
• This approach facilitates more precise and personalized treatment planning, potentially improving patient outcomes.
• The study highlights the potential of advanced deep learning architectures in revolutionizing radiotherapy.