Abstract: The purpose of this research is to provide digital diagnostic templates as references for automatic landmark identification and attempt to realize landmark automatic identification by computers. 2249 pre-treatment X-ray films of malocclusion patients were divided into 21 subtypes according to the coordinates of 60 landmarks by cluster and discriminate analysis. The total differentiate rate and leave-one-out differentiate rate were 89.1% and 85.0% respectively. 21 digital diagnostic templates were established. A new case could be classified into one subtype by discriminate functions or the characters of templates and the digital diagnostic templates could be used for diagnosis, evaluation and prognosis in orthodontic clinic. 23 landmarks of 10 new samples were identified automatically by computer using templates. The mean errors of 11 landmarks were below 2 mm, which could reach the clinical demand.

Key words: digital diagnostic templates, cluster and discriminate analysis, cephalometrics, automatic identification

摘要： 为了给计算机自动识别诊断提供模板参考并初步尝试计算机自动识别, 将2249例错牙合畸形样本进行聚类和判别分析, 以60个标志点的坐标值作为分类变量, 形成21个数字化诊断模板, 总判别准确率和交互验证准确率分别达到89.1%和85.0%。采用判别方程或者模板特征对新样本进行分类, 并为正畸临床诊断、疗效评价和预测提供参考。 采用模板匹配的方法对10例新样本的23个标志点进行初步计算机自动识别研究, 其中11个标志点的识别误差小于2 mm, 能够满足临床应用要求。

关键词: 数字化诊断模板, 聚类和判别分析, X线头影测量技术, 自动识别