Research On Key Technologies In Three-Dimensional Image-Guided Brachytherapy For Cervical Cancer

Cervical cancer is the most common malignant tumor of the female reproductive system.Brachytherapy plays an important role in the treatment of cervical cancer.Especially for the treatment of locally advanced cervical cancer,brachytherapy is an essential part of the standard therapy and closely associated with improvements in clinical outcomes.The emergence and application of 3D image-guided brachytherapy are significant technical advancements for the treatment of cervical cancer.In order to achieve more accurate,efficient and optimized treatment,researches are carried out with the focuses on key technologies of 3D image-guided brachytherapy for cervical cancer including applicator design and application,contour delineation,applicator reconstruction and treatment plan optimization.The main contributions of the thesis are as follows:(1)The methods to achieve individualization and dose-based optimization of the needle configuration for combined intracavitary-interstitial brachytherapy of cervical cancer are proposed.Virtual design and dosimetry evaluation for the interstitial needle configuration are achieved.Utilizing the constructed mixed integer linear programming model,the needle configuration and corresponding dwell times can be simultaneously optimized according to the dosimetric objectives.The treatment planning study shows that superior dosimetric results can be obtained with the proposed methods than the use of standardized applicator products for the locally advanced cervical cancer patients with challenging tumors or extra lateral expansion.(2)The automatic segmentation technique based on convolutional neural networks is first applied in the treatment planning of image-guided brachytherapy for cervical cancer.A novel network architecture DSD-UNET is proposed and used to achieve automatic segmentations of target volume and organs at risk in the planning CT images.Real-world patient data is collected to train and test the neural networks.The testing results for the trained models show that the proposed DSD-UNET model outperforms the classic 3D U-Net model on segmentations of all anatomical structures.The DSDUNET model achieves relatively good segmentation accuracy for the target volume,bladder and rectum.The proposed automatic segmentation method could be useful to improve the efficiency and consistency of treatment planning for cervical cancer brachytherapy.(3)The method of fully automatic applicator reconstruction based on convolutional neural networks is proposed.The applicator reconstruction is achieved automatically with DSD-UNET based segmentations of applicator components followed by 3D skeletonization and polynomial curve fitting.The testing result for the automatic reconstruction of tandem and ovoid applicator in the planning CT images shows that the proposed method can achieve automatic applicator reconstruction with high accuracy.The proposed method could be useful to improve the accuracy and efficiency of treatment planning for cervical cancer brachytherapy.(4)The inverse treatment planning method based on the equivalent uniform dose is proposed,then the optimization objectives of equivalent uniform dose can be applied to the treatment plan optimization for image-guided brachytherapy of cervical cancer.The results of treatment planning and dosimetric studies show that the application of equivalent uniform dose based optimization objectives can obtain better dosimetric results compared to the inverse treatment planning based on physical dose objectives alone.The proposed method is a more flexible and effective inverse treatment planning approach for image-guided brachytherapy of cervical cancer.