Research On Rectal Toxicity Prediction For Cervical Cancer Radiotherapy

Cervical cancer is the third most common cause of female cancer mortality worldwide.The combination of the external beam radiotherapy(EBRT)and brachytherapy(BT)(EBRT+BT)is a common therapy regime for locally advanced cervical cancer.Boosting higher dose is an effective means to improve the local control rate in cervical cancer radiotherapy.However,a high dose may substantially increase toxicity risks to nearby organs at risk(OARs),such as the rectum,sigmoid,bladder,and vagina.Particularly,three volumetric dosimetric parameters D0.1cc,D1cc and D2cc were used to establish dose-toxicity relationship in the occurrence of rectal morbidity.But the issue with current D0.1/1/2cc evaluation procedure is that the cumulative dose is summed with an assumption that the hotspot regions are stationary throughout the entire fractional treatments.However,this static assumption is often violated by the large inter-fraction rectum deformation,especially in intra-cavity brachytherapy treatment cases,which will cause the current D0.1/1/2cc are not consistent with the real dose.Moreover,D0.1/1/2cc are inherently in deficiency of dosimetric spatial information.Better understanding of the relationship between OAR toxicity and dose is critical for safe dose escalation to improve local control of large-size advanced-stage cervical cancer tumors.Taking this as the starting point,we carried out a series of related researches.In this paper,some basic principle algorithms related to our study were detailed firstly,including a previously developed local topology preserved non-rigid registration point matching algorithm(TOP-DIR)algorithm and the logistic regression model and support vector machine model(SVM)in machine learning,the convolution neural networks(CNN)in deep learning.To analyze the relationship between rectal toxicity and dose in cervical radiotherapy cancer,two rectal toxicity prediction models are proposed and implemented as following:Firstly,the machine learning based rectal toxicity prediction was proposed.In this study,the accumulated equivalent 2-Gy rectal surface dose was deformably summed using the deformation vector fields obtained through TOP-DIR point matching algorithm.and then the cumulative three-dimensional(3D)dose was flattened and mapped to a two-dimensional(2D)plane to obtain the rectum surface dose map(RSDM).The dose volume parameters(DVPs)were calculated from the 3D rectum surface dose,while the texture features and the dose geometric parameters(DGPs)were extracted from the 2D RSDM.Representative features selected by sequential feature selection(SFS)procedure were feed into SVM.The satisfactory predictive accuracy of sensitivity 84.75%,specificity 79.87%,and AUC 0.91 show that the rectal toxicity prediction model which owns high predictive accuracy can provide a reliable support for cervical cancer patients.Secondly,the deep learning based rectal toxicity prediction was proposed.In this study,2D RSDMs as the image data were obtained via 3D-2D dose mapping procedure on the deformably accumulated rectal surface dose.We adopted a transfer learning strategy to overcome the limited patient data issue.A 16-layers CNN(VGG-16-CNN)was pre-trained on a large-scale natural image database,ImageNet,and fine-tuned with RSDMs.The gradient weighted class activation maps(Grad-CAM)were also generated to highlight the discriminative regions on the RSDM along with the prediction model.The satisfactory predictive accuracy of sensitivity 75%,specificity 83.3%,and AUC 0.89 on leave-one-out cross-validation show that transfer learning strategy can successfully solve the problem of our limited image data.Besides,VGG-16-CNN also show its powerful superiority to build high-performance rectal toxicity prediction model.And the relationship between rectal toxicity and dose distribution could be better understood.In this paper,two rectal toxicity prediction models were proposed by using machine learning algorithm with the features extraction based on the prior knowledge and deep learning algorithm with automatic features extraction for the evaluation of rectum toxicity prediction for cervical cancer radiotherapy.The extensive experimental results have demonstrated the feasibility of the proposed scheme for rectal toxicity prediction,rendering it a potential tool for analyzing the relationship of the rectum toxicity and dose.Though some preliminary achievements in the study have been obtained in this paper,there are still many defects and deficiencies that need to be improved and solved,such as limited image database,the machine learning based model only considers the dose features,the deep learning based model lacks the comparison with different CNN algorithms.Therefore,it still needs to be further explored and perfected on the current research.