基于锥形线束CT数据的智能颈椎骨龄评估系统的建立

目的：建立智能颈椎骨龄评估系统，初步评估基于锥形线束CT数据的智能颈椎骨龄评估系统的可靠性和临床应用价值。方法：选取60例生长发育期（8~16岁）儿童同时段拍摄的侧位体层片和锥形线束CT作为实验数据。在锥形线束CT上通过Otsu算法提取患者的面部区域，使用三维最小二乘法获得一个矢状面，并在此矢状面上应用超像素算法来对图像进行分割以获取颈椎区域，随后分别进行人工定点与形态学算法的自动定点，对两组坐标数据进行一致性检验。根据颈椎骨龄分期指南的定义，进行算法设计，建立智能颈椎骨龄评估系统。同时通过同期的侧位体层片进行人工颈椎骨龄判读。采用加权Kappa一致性检验及Gamma相关度检验比较人工侧位体层片颈椎骨龄判读结果与智能颈椎骨龄评估结果，判断智能颈椎骨龄预测系统的临床应用价值。结果：基于锥形线束CT数据自动化捕捉的颈椎形态整体上具有较高的形态识别度，在预测13个点中的8个拐点时，自动定点与人工定点在X轴和Y轴上的Wilcoxon检验结果差异均无统计学意义（均P>0.05）。同时，智能系统的颈椎骨龄评估结果与人工识别结果有较强的一致性和相关性（加权Kappa值0.877，Gamma值0.991，均P<0.05）。结论：基于锥形线束CT数据进行的自动化颈椎形态捕捉和智能颈椎骨龄预测系统有一定的可靠性，自动化程度高，具有一定的临床应用价值。

Establishment of an intelligent cervical vertebrae maturity assessment system based on cone beam CT data

Objective:To establish an intelligent cervical vertebra maturity assessment system, and to evaluate the reliability and clinical value of the system. Methods: Sixty children aged 8-16?years were recruited in the study. Lateral cephalometric radiograph and cone beam CT (CBCT) were taken at the same period. Based on the CBCT data, the system automatically extracted the patient’s facial area through Otsu’s method, intercepted the sagittal plane by three-dimensional least squares method, captured the second to fourth cervical vertebrae by superpixel segmentation. And then selected points were marked automatically through morphological algorithm and manual method. Consistency test was performed on the two sets of data to compare the reliability of automated cervical morphology capture. According to the parameters of morphological identification, positioning and staging algorithms were designed to form the intelligent cervical vertebra maturity assessment system. The cervical vertebra maturity was also judged manually on the lateral cephalometric radiograph. The weighted Kappa test and the Gamma correlation coefficient were subsequently applied to evaluate the consistency and correlation. Results: The results showed that the cervical vertebra features automatically captured based on CBCT data had a high accuracy on the overall morphological recognition. In the prediction of 8 inflection points out of 13 points, there was no significant difference between automatic and manual method on both X and Y axes (all P>0.05). The assessment results of the cervical vertebra maturity of the intelligent system had strong consistency and correlation with the manual recognition results (weighted Kappa value=0.877, Gamma value=0.991, bothP<0.05).Conclusion:The intelligent cervical vertebrae maturity assessment system based on CBCT data established in this study presents reliable outcome and high degree of automation, indicating that the system may be used clinically.