基于结构相似性指数评估放疗计划质量的可行性研究

目的 利用结构相似性指数（structural similarity index，SSIM）对放疗计划的三维剂量分布进行评价，比较SSIM相对现有常用质量评价指数的优缺点，同时构建以SSIM为基础的多处方剂量计划的评估方法。方法 引入SSIM来分析实际与理想三维剂量数据并结合其空间位置信息给出各体素质量评分，进而计算感兴趣区域（ROI）内的平均值作为该区域的质量评分。选取2018年1月至2020年5月北京大学肿瘤医院既往已实施治疗的53例临床同步推量放疗计划，抽取两例计划分别引入两种质量缺陷（局部高量、剂量跌落）与正常计划进行对比实验来表征SSIM指数的响应，展示SSIM指数评价结果的图像模式。分析53例放疗计划各ROI的均匀性指数（HI）、适形性指数（CI）与SSIM的相关性，评估SSIM反映均匀性和适形性的能力。结果 SSIM指数的图像模式能够反馈剂量分布存在质量缺陷的位置信息。HI与SSIM在转移与可疑淋巴结（PGTVnd）中由于HI敏感性降低导致未表现出相关性，在计划靶区（PTV）去除PGTVnd后的区域中存在显著负相关（R=-0.86，P<0.01），CI与SSIM指数在PGTVnd中存在显著正相关（R=0.83，P<0.01）。结论 SSIM指数在改善以往评价参数缺陷的基础上，能够结合均匀性与适形性对计划进行质量评价，同时精准反馈存在质量缺陷的空间位置。

A feasibility study of the application of structural similarity index to the assessment of radiotherapy plan quality

Objective To assess the three-dimensional dose distribution in radiotherapy plans using the structural similarity index(SSIM), compare the performance of SSIM with commonly used quality assessment indices, and develop a SSIM-based quality assessment method of multiple prescribed doses. Methods The SSIM was introduced to providea quality score of various voxels by comparing actual and ideal three-dimensional dose data and combining the spatial location information of the voxels. Then the average value in a region of interest (ROI) was calculated as the quality score of the region. Fifty-three cases of cervical cancer were selected to analyze the correlation of the SSIM with the uniformity index (HI), conformity index (CI) of the dose distribution in various ROIs and to explore the capability of the SSIM to reflect the uniformity and conformity of dose distribution.Two types of quality defects were individually introduced into two of 53 radiotherapy plans. Then the two plans were compared with normal plans to characterize the response of the SSIM. Results There was no correlation between HI and SSIM in positive lymph nodes(PGTVnd) due to the decrease in the HI sensitivity, while there was a significant negative correlation between them in regions where PGTVnd was removed from the planning target volume(PTV, R=-0.86, P<0.01). Meanwhile, there was a significant positive correlation between CI and SSIM in PGTVnd (R=0.83, P<0.01). Therefore, the SSIM can be used to identify the artificial design defects in plans by determining abnormal dose gradients. Conclusions Apart from reducing the defects of previous assessment parameters, the SSIM has the capability to assess the quality of radiotherapy plans by combining the uniformity and conformity of dose distribution and can provide accurate feedback on the spatial locations of quality defects.