2888 - A Local-Global Integration Network for Enhancing CBCT Quality in Cervical Cancer Radiotherapy

10:45am - 12:00pm PT

Hall F

Screen: 8

POSTER

Presenter(s)

Hua Yang, MD - The First Affiliated Hospital of Air Force Medical University, Xi’an, Shaan’xi

H. Yang¹, D. Huang², Y. Zhang³, W. Li¹, L. C. Wei⁴, and L. N. Zhao⁵; ¹Department of radiation oncology, Xijing Hospital, the Fourth Military Medical University, Xi'an, China, ²Department of Military Biomedical Engineering, Air Force Medical University, Xi’an, Shaanxi, China, ³Department of Radiation Oncology, First Affiliated Hospital of Air Force Medical University, Xi'an, China, ⁴Department of Radiotherapy, The First Affiliated Hospital,the Air Force Medical University, Xi’an, China, ⁵Department of Radiation Oncology, Xijing Hospital, Fourth Military Medical University, Xi'an, Shaanxi, China

Purpose/Objective(s): The purpose of this study is to address the limitations of Cone Beam Computed Tomography (CBCT) in cervical cancer radiotherapy, specifically focusing on improving image quality for better targeting and implementation of Adaptive Radiation Therapy (ART). Current CBCT-to-CT conversion methods often neglect the critical local cervical region and suffer from imperfect matching due to cervical deformation. This study aims to develop a novel Local-Global Integration Network to enhance both global and local image quality.

Materials/Methods:

We propose a Local-Global Integration Network consisting of two subnetworks: a Local Generation Subnetwork with a Local Fusion Module and a Global Discrimination Subnetwork. The Local Fusion Module focuses on improving the quality of the cervical region by fusing local features into the global feature map. The Global Discrimination Subnetwork evaluates the entire image by dividing it into multiple blocks for detailed discrimination. The network was trained using paired CBCT and CT images from 20 cervical cancer patients, resulting in 1362 image pairs. Evaluation metrics included peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), mean absolute error (MAE), and a comprehensive image quality index (CIQI) that combines local and global evaluations.

Results:

Experimental results demonstrate significant improvements in both local and global image quality. The proposed model achieved SSIM values of 71.14% ± 1.21 for the local region and 76.07% ± 2.12 for the global region, outperforming models without the Local Fusion Module or alternative discriminative designs. Additionally, the CIQI metric showed a marked improvement, increasing from 29.30 ± 2.32 (baseline) to 56.80 ± 2.10. Visualization comparisons further confirmed enhanced details in tumor contours and surrounding structures.

Conclusion:

The proposed Local-Global Integration Network effectively improves CBCT image quality for cervical cancer radiotherapy by emphasizing the local cervical region while maintaining global detail. This approach contributes to more precise and effective treatment planning, paving the way for successful implementation of Adaptive Radiation Therapy in clinical practice.