Orthognathic Surgical planning based on Soft Computing Science including Self Organizing Map

我们开发了一种基于软计算的人工智能系统，用于颌骨畸形矫正手术中正颌手术的规划。自组织映射神经网络（SOM）具有从高维空间映射到低维空间并保持其拓扑结构的特点。因此，类似的案例在地图上随处可见。利用这一特征，我们将SOM应用于颌骨畸形软组织轮廓的分类。资料为44例患者术前、术后侧位脑电图。测量软组织轮廓并将其分为7组。选取11个软组织测量值作为SOM的输入向量。采用遗传算法对各测量参数进行优化。基于SOM和遗传算法的分层分类方法与先前的剖面诊断结果的对应率约为100%。在规划正颌手术病例时，参考与新病例相似的既往文献是有意义的。我们构建了数据库软件，可以从SOM存储的大量不同案例中自动搜索相似的案例。它也有助于预测手术后的轮廓。轮廓变化是骨骼运动和软组织厚度相互作用的非线性结果。使用新的模糊神经元来近似输入和输出之间的非线性关系。比较新模糊神经元法与线性多元回归法的预测误差，新模糊神经元法适用于正颌手术后结果轮廓的预测。利用变形技术，可以在手术前对患者的预测轮廓进行拍照。

We developed an artificial intelligent system based on soft computing for planning of orthognathic surgery in surgical correction of jaw deformities.The self-organizing map neural network (SOM) has a feature of a mapping from a high dimensional space to a low dimensional space with keeping its topology. Accordingly, similar cases are closely located on the map. Using this feature, we applied SOM to classification of soft-tissue profiles of jaw deformities. Materials were pre-and post-operative lateral cephalograms of 44 patients. Soft-tissue profiles were measured and classified (diagnosed) into seven groups. Eleven soft-tissue measurements were selected as input vector of SOM. The genetic algorithm (GA) was employed to optimize coefficients for each measurement. The hierarchical classification method based SOM and GA resulted about 100% corresponding to the previous profile diagnosis.In planning of orthognathic surgery case, it is meaningful to refer the past documentation similar to the new case. We constructed database software that automatically searches for similar cases from a number of various cases stored by means of SOM. It is also useful to predict the profile after surgery. Profile change is a nonlinear result from the interaction of not only skeletal movement but also soft tissue thickness. Using neo fuzzy neuron that can approximate a nonlinear relationship between inputs and outputs. Comparing prediction errors between the neo frizzy neuron method and the linear multiple regression method, the neo fuzzy neuron method was suitable for prediction of resultant profile after orthognathic surgery. A photograph of the patient's predicted profile can be made from that before surgery using morphing technique.