基于人工神经网络的牙本质瓷计算机配色研究

Matching the color of a natural tooth to a ceramic restoration is one of the most challenging aspects in esthetic prosthodontics. Although advances in shade guides and tooth color measurement devices have already made clinical shade selection more accurate and objective, layering the pocerlain is still a subjective process which is affected by many factors. To eliminate the subjective aspect of pocerlain layering, computer color matching has been introduced into dentistry. Computer color matching is based on Kubelka-Munk theory and has been widely used in dyeing and printing industry. But there were some problems when the theory was applied in color matching in dentistry. In fact, there was no linear correlation between percelain powder and color coordinates. In this study, we introduce Back Propagation Neural Network into the field of computer color matching in dentistry. Different proportion of dentin porcelain powders were mixed and fabricated into ceramic dics ,their color coordinates were measured, then BP neural network was used to establish the nonlinear correlation between the ingredient of the dentin porcelain and the color. The software program will be developed and trained to calculate the prescription of porcelain powder according to specific color coordinates, which will improve the color matching in restorative dentistry.

牙齿颜色的识别与再现是瓷修复体制作的关键性问题，比色板的改进和比色仪.器准确性的提高使临床选色更精确和客观，但技师制作修复体的过程仍然是主观过程，受多.种因素的影响。计算机配色的应用为上述问题提供了新的解决途径，以 Kubelka-Munk 理论为理论基础的传统计算机配色方法广泛应用于印染工业，但不完全适用于瓷修复体的配色,并且在应用中有大量复杂的矩阵计算，需进一步改进。基于瓷粉的成分与瓷块色度值之间为非线性关系，本研究将处理非线性问题的有效工具 BP 神经网络引入修复体的计算机配色中，通过配制不同成分牙本质瓷粉，制作瓷试件，测量其色度值，并通过实验确定神经网络结构，对神经网络进行训练、测试和改进，建立基于人工神经网络的牙本质瓷粉配色系统，实现牙本质瓷粉成分与试件色度值的映射模拟，以期进一步提高牙本质瓷粉的配色精度。为实现修复体更复杂的计算机配色做前瞻性研究，使修复体的制作更客观和精确。

烤瓷修复体的制作目前主要依赖技师，是一个主观的过程。牙本质瓷的颜色主要决定了整个修复体的颜色。本项目将BP人工神经网络应用于牙本质瓷计算机配色研究中，通过建立颜色与瓷粉成分的非线性映射，当测量出牙齿颜色数据时，即可将颜色数据输入本系统预测出烤瓷冠颈、中、切三部分的牙本质瓷粉成分，这样就实现牙本质瓷粉配色的自动化与客观化。本研究主要工作内容有：建立天然牙颜色数据库及牙本质瓷块颜色数据库，通过分析比较发现牙本质瓷块的颜色与天然牙颜色较好符合，为临床应用奠定基础；通过实验计算不同结构的BP神经网络总误差，确定了最佳的神经网络隐含层数、隐含层节点数、输入层到隐含层的激励函数及隐含层到输出层的激励函数，从而实现了BP神经网络的构建，并用Java语言编写为相应程序方便应用，临床实验证明牙本质瓷颜色预测系统用于烤瓷冠制作可达到与视觉比色法及仪器比色法类似的效果，为将来烤瓷冠制作的客观化打下基础。