Study On Precise Delineation Of Organs At Risk Of Radiotherapy Based On An Automatic Contouring Software And Chinese Digitized Visible Human

Background and purposeAt present,cancer is still a major disease threatening our life,health and family happiness.With the continuous progress of science and technology,the diagnosis,treatment,rehabilitation and other aspects of cancer treatment in the medical field are also developing more efficiently and accurately,however,the traditional radiotherapy contour delineation is manual,time-consuming and laborious,and there are great individual differences in accuracy and repeatability.This research is originated in the National Key Research & Development Program,with a self developed deep learning-based radiotherapy automatic contour software which integrates high-precision Chinese digital visible human models,to study on precise,efficient and automatic delineation of organs at risk of radiotherapy.The purpose of this study is to explore the automatic and intelligent technology of radiotherapy contour delineation,so as to liberate the labor force of radiotherapy doctors,improve the efficiency and the accuracy of radiotherapy,shorten the waiting time of radiotherapy patients,and provide intelligent software equipment and ideas for the implementation of homogeneous radiotherapy in different levels of hospitals.MethodsPart one: Research and development of a radiotherapy contouring software based on deep learningDevelop the software basic architecture with C ++ language,and using the deep learning neural network technology to develop the automatic outlining function of the radiotherapy contouring software,and then verification and validation of the software.Part two: a comparative study on the Yorktal-CS for delineating organs at risk in intensity-modulated radiotherapy for nasopharyngeal carcinomaA total of 100 NPC patients who underwent IMRT in five hospitals of different levels in Chongqing from January 1,2016 to December 31,2018 were included,with the cases and imaging data collected.CT images of 12 OARs of eight kinds(bilateral eyes,bilateral lens,bilateral optic nerves,bilateral parotid glands,the oral cavity,the spinal cord,the brain stem and the brain)were delineated using Yorktal-CS(automatic delineation group),while 2commercially available TPSs,Eclipse(control group 1)and Fonics Plan(control group 2).Time elapsed during delineation was recorded for comparison and the quality of automatic delineation was analyzed by Dice Similarity Coefficient(DSC).Data were analyzed using SPSS 25.0 Statistical software.Friedman rank-sum test was used for the comparison among three samples.In cases when significant difference exists among groups,Dunn Bonferroni test was used for post-hoc comparison of each two groups.Wilcoxon rank-sum test was used for comparison between control groups 1 and 2.All statistical analyses were conducted with the two-sided probability test.Test level α=0.05 and P<0.05 was defined as statistically significant.Part three: Research on precise outline of hippocampus and some other organs at risk in head and neck based on Chinese Digitized Visible HumanA total of 30 head and neck radiotherapy patients in The Second Affiliated Hospital of Army Medical University in Chongqing from January 1,2018 to December 31,2018 were included,with the cases and imaging data collected.The one the shape of the most similar form four sets of Chinese Digitized Visible Human was selected to register and fuse to the individual patient’s brain CT image by Yorktal-CS automatic registration and fusion function,and the hippocampus area and some other organs at risk in head and neck were accurately observed and delineated according to the deformed Chinese Digitized Visible Human.Show and evaluate the effect of registration and fusion of hippocampal delineation and some other organs at risk in head and neck.ResultsPart one: Research and development of a radiotherapy contouring software based on deep learningCompleted the research and development.Obtained the physical software with the performance improving.Yorktal-CS software,based on medical imaging technology,has automatic segmentation technology of OARs and three-dimensional reconstruction technology of organs and tissues.It is composed of five modules.In addition to the conventional delineation function,it loads Chinese digital visible human model,displays the radiation contour in three dimensions,automatically outlines 17 organs with one key,and has good data compatibility transfer and can edit files between some kinds of domestic and imported TPSs.Part two: a comparative study on the Yorktal-CS for delineating organs at risk in intensity-modulated radiotherapy for nasopharyngeal carcinomaComparison of delineation speed of OARs: Three kinds of software were used to delineated 12 OARs in NPC patients independently.The delineation time of Yorktal-CS was225.33(222.02-228.86)seconds,compared with 1081.31(1056.92-1094.49)and 1945.80(1891.97-1996.96)seconds in Eclipse and Fonics Plan,respectively.The difference among three groups were statistically significant(x2=592.62,P=0.000),as well as that between each two groups(all P-values were 0.000).Meanwhile,compared with Eclipse TPS,the Fonics Plan required longer delineation time,and significant statistical difference also exists in each OARs.Evaluation of DSC values: Yorktal-CS was used for automatic delineation.The median DSC values of OARs were greater than 0.8 except that in bilateral optic nerves(left,0.76;right,0.77)and the left lens(0.79).The maximum DSC value in all outlined OARs was 0.94 found in bilateral eyes,while the minimum value was 0.66 found in the oral cavity.The DSC values of bilateral eyeballs ranged from 0.85 to 0.94,while those of bilateral optic nerves ranged from 0.74 to 0.81.Interquartile range: bilateral eyes is 0.03 and the oral cavity is 0.13.Part three: Research on precise outline of hippocampus and some other organs at risk in head and neck based on Chinese Digitized Visible HumanIn this study,head and neck CT images of 30 patients were registered and fused,and 30 cases were successful and 0 case was failed.The display and outline effects of brain stem,pharyngeal constrictor,parotid gland,optic chiasm and hippocampus which were registered and fused by CVH were demonstrated.ConclusionBased on the deep learning technology,the independently developed radiotherapy contour delineation software(Yorktal-CS)organically integrates the high-precision Chinese digital visible human models and can better match and fuse with the head and neck CT of individual patients.In the outlining of most risk organs in nasopharyngeal carcinoma radiotherapy,the contouring speed is much higher than that of the two mainstream commercial radiotherapy software,meanwhile the accuracy is ensured.