机器学习在口腔植入体无损测量中的应用

Application of Machine Learning for Nondestructive Measurement of Oral Implants

目的为解决患者就诊时档案丢失无法确定口腔内种植体规格的问题,探索一种针对口腔植入体规格的无损测量方法。方法首先,对口腔曲面CT图像使用分段阈值、图像分割、形态学处理等方法进行预处理。接着,将处理后的图像提取其灰度共生矩阵和梯度方向柱状图两种纹理特征用以表示口腔植入体的特点。然后,比较8种高效的分类器算法,选择分类表现最优的基于灰度共生矩阵的RBFSVM方法用以检测植入体的位置。最后,根据检测结果计算植入体的实际规格参数。结果本研究共使用420张口腔CT图像进行实验,最终结果与植入体实际规格的平均误差<2%。结论本研究提出的机器学习在口腔植入体无损测量方法显示出较好的测量精度,有效帮助医生确定病人口腔内已有的植入体的规格。

Objective Introduce a non-destructive measuring method for oral implants features,in order to solve the problem that the implants specification cannot be determined when the patient’s file is lost.It also provides a method and idea for the non-destructive measurement of various implants(such as bone nail,bone plate,etc.).Methods First,segmentation threshold,morphological processing and morphological segmentation were used for preprocessing.Then,GLCM and HOG were used to represent the characteristics of implants.And the RBFSVM with the best classification performance was selected to detect the position of the oral implants after comparing 8 efficient classifiers.Finally,calculate the physical size of the implants based on the result and achieved the nondestructive measurement of oral implants.Results A total of 420 images were used in this project,and the average error between the results and the actual size of implants was less than 2%.Conclusion It shows a good measurement accuracy that can help doctors to determine the specification of the implants in the patient’s mouth.