李彩虹, 王沛沛, 李金凯, 曹远东, 孙新臣. 基于Python实现鼻咽癌放疗计划剂量学指标的自动提取和量化评价[J]. 国际放射医学核医学杂志. DOI: 10.3760/cma.j.cn121381-202303018-00404
引用本文: 李彩虹, 王沛沛, 李金凯, 曹远东, 孙新臣. 基于Python实现鼻咽癌放疗计划剂量学指标的自动提取和量化评价[J]. 国际放射医学核医学杂志. DOI: 10.3760/cma.j.cn121381-202303018-00404
Caihong Li, Peipei Wang, Jinkai Li, Yuandong Cao, Xinchen Sun. Application of Python in quantitative evaluation of radiotherapy plan for nasopharyngeal carcinoma[J]. Int J Radiat Med Nucl Med. DOI: 10.3760/cma.j.cn121381-202303018-00404
Citation: Caihong Li, Peipei Wang, Jinkai Li, Yuandong Cao, Xinchen Sun. Application of Python in quantitative evaluation of radiotherapy plan for nasopharyngeal carcinoma[J]. Int J Radiat Med Nucl Med. DOI: 10.3760/cma.j.cn121381-202303018-00404

基于Python实现鼻咽癌放疗计划剂量学指标的自动提取和量化评价

Application of Python in quantitative evaluation of radiotherapy plan for nasopharyngeal carcinoma

  • 摘要:
    目的 利用Python和DICOM RT文件实现鼻咽癌放疗计划剂量学指标的自动提取和量化评价,为临床及科研实践提供参考。
    方法 回顾性分析2022年1至10月在江苏省人民医院进行放疗的21例鼻咽癌患者的临床资料,其中男性13例、女性8例,年龄(48.7±21.6)岁。在Monaco治疗计划系统中采用人工方法统计每例患者各放疗靶区的最大剂量(Dmax)、最小剂量(Dmin)、平均剂量(Dmean),覆盖靶区98%、50%、2%体积的最小剂量(D98%、D50%、D2%),95%、100%、105%、110%处方剂量覆盖的靶区体积占其总体积的百分比(V95%、V100%、V105%、V110%),危及器官1 cm3、1%体积的最小剂量(D1cc、D1%)、Dmax、Dmean和30 Gy剂量覆盖的器官体积占其总体积的百分比(V30 Gy),并利用公式计算适形指数(CI)、均匀性指数(HI)和梯度指数(GI),然后导出RT Structure和RT Dose文件,由Python方法实现放疗计划剂量学指标的自动提取和量化评价,并与人工量化结果进行对比分析。2种方法的比较采用配对样本t检验(方差齐)。
    结果 Python方法获取到的各靶区的Dmin、Dmean、Dmax、D98%、D50%、D2%、V95%、V100%、V105%、V110%和HI等剂量学指标,与人工方法相比,差异均<3%。其中,PTV1的Dmin差异最大,为2.21%±0.17%;PTV2的Dmin次之,为1.75%±0.13%;PGTVnd的V105%为1.67%±0.09%,二者间的差异无统计学意义(t=0.093、0.071、0.065,均P>0.05)。CI和GI的最大差值分别为0.007±0.108和0.016±3.958,差异无统计学意义(t=0.011、0.034,均P>0.05),但Python方法可计算出S-index和HI3。危及器官评价结果显示,Python方法统计的视神经和视交叉的Dmax较人工方法略低,差值>1 Gy,且差异均有统计学意义(t=−2.909、−3.420,均P<0.05),其余评价结果基本一致,差值均<0.5 Gy或0.5%。在放疗计划评价用时方面,单个病例Python方法仅需1~3 min,较人工方法的30~50 min明显缩短。
    结论 Python方法自动提取到的剂量学数据与人工方法基本一致,但过程用时大幅缩短,量化评价指标更全面,具有普适性,能为后续的基于人工智能的放疗计划自动评价提供技术参考。

     

    Abstract:
    Objective To realize the automatic extraction and quantitative evaluation of dosimetric indicators of nasopharyngeal carcinoma radiotherapy plan to provide reference for clinical and scientific research practice with Python and DICOM RT files.
    Methods The clinical data of 21 patients with nasopharyngeal carcinoma who underwent radiotherapy in Jiangsu Province Hospital from January 2022 to October 2022 were retrospectively analyzed. Among them, 13 cases were males and 8 cases were females, aged (48.7±21.6) years. Firstly, the maximum dose(Dmax), the minimum dose(Dmin), the mean dose (Dmean), the dose covering 98%, 50%, 2% of the volume(D98%, D50%, D2%), the volume covered by 95%, 100%, 105%, 110% prescription dose(V95%, V100%, V105%, V110%)of each target and the dose covering 1 cc, 1%(D1 cc, D1%), Dmax, Dmean and V30 Gy of the organs at risk were manually counted in the Monaco system, and the conformity index(CI), homogeneity index(HI) and gradient index(GI) were calculated by the formula. Secondly, RT Structure and RT Dose files were exported, and the same automatic extraction and quantitative evaluation of radiotherapy plans were completed through Python, compared with the results of manual quantification. Data from the two groups that fit a normal distribution were compared by paired t-test.
    Results The difference of Dmin, Dmean, Dmax, D98%, D50%, D2%, V95%, V100%, V105%, V110% and HI of each target region obtained by Python method and manual method was less than 3%. The Dmin difference of PTV1 was the largest, which was 2.21%±0.17%; The Dmin difference of PTV2 was the second, 1.75%±0.13%; The V105% difference of PGTVnd was the third, 1.67%±0.09%, and the difference was not statistically significant (t=0.093, 0.071, 0.065, all P>0.05). The maximum difference of CI and GI was 0.007±0.108 and 0.016±3.958, and the difference was not statistically significant (t=0.011, 0.034, both P>0.05), but Python method could calculate S-index and HI3. Regarding organs at risk, the Dmax of Optic Nervus and Chiasma was slightly lower by Python method than by manual method, and the differences greater than 1 Gy were statistically significant(t=−2.909, −3.420, both P<0.05). In addition, the other quantitative evaluation results of the organs at risk obtained by the two methods were nearly the same, and the difference < 0.5 Gy or 0.5%. In terms of time consumption, Python method only takes 1~3 min, which was significantly shorter than 30~50 min of manual method.
    Conclusions The dosimetric data automatically extracted by the Python method which has significantly reduced time consumption, more comprehensive evaluation indicators, and universality is basically consistent with the manual method, which can provide technical reference for subsequent automatic evaluation of radiotherapy plans based on artificial intelligence.

     

/

返回文章
返回