基于锥形束CT的牙齿髓腔分割算法研究

发布时间：2019-05-27 19:01

【摘要】：最新的医学研究发现牙髓腔沉积可用于法医学中的人体年龄推算,基于目前流行的口腔CBCT(Cone Beam CT,锥形束CT)成像技术,如何准确实现牙髓腔的三维断层图像分割是应用的前提。由于CBCT牙齿图像存在噪声干扰、牙齿界限模糊、牙齿与牙槽骨灰度值相近等因素的影响,使得牙髓腔的准确分割存在较多的难点。PCNN(Pulse Coupled Neural Network,脉冲耦合神经网络)具备生物学背景,能够从复杂背景下提取有效信息,具有同步脉冲发放和全局耦合等特性,其信号形式和处理机制更符合人类视觉神经系统的生理学基础。本论文在PCNN理论与应用深入研究的基础上,提出了一种改进的PCNN模型,实现了 CBCT牙髓腔的三维断层序列图像的准确分割。论文完成的主要工作和创新点有:(1)针对传统的PCNN模型结构繁冗、大量人工设定参数、阈值衰减不稳定等问题,本文通过调整PCNN接受部分网络结构,在保证其生物学特性的前提下,提出了一种改进的PCNN模型,优化了神经元外部输入、连接输入L权重取值和阈值衰减方式,实验结果证实该模型有效地降低了算法的复杂度,提高了像素空间信息的描述能力。(2)针对牙齿图像的模糊性导致PCNN模型迭代次数难以确定且需要人工设置的问题。本文结合了图像像素点的空间信息,分析了迭代次数对PCNN模型分割算法的影响,利用信息熵最优化准则,给出了一种基于最小交叉熵的最佳迭代次数判定准则,实现了改进型PCNN模型分割算法对CBCT牙齿髓腔图像的准确分割。
[Abstract]:The latest medical research has found that dental pulp deposition can be used to calculate the age of human body in forensic science, based on the popular oral CBCT (Cone Beam CT, conical beam CT imaging technology. How to accurately realize the three-dimensional tomographic image segmentation of pulp cavity is the premise of application. Due to the influence of noise interference, blurred tooth boundary and similar bone gray value between teeth and alveolar bone, there are many difficulties in the accurate segmentation of pulp cavity. PCNN (Pulse Coupled Neural Network, Pulse coupled neural network (PNN) has biological background, can extract effective information from complex background, has the characteristics of synchronous pulse release and global coupling, and its signal form and processing mechanism are more in line with the physiological basis of human visual nervous system. In this paper, based on the in-depth study of PCNN theory and application, an improved PCNN model is proposed to realize the accurate segmentation of three-dimensional tomographic sequence images of CBCT pulp cavity. The main work and innovations of this paper are as follows: (1) aiming at the complicated structure of the traditional PCNN model, a large number of manual setting parameters, unstable threshold attenuation and so on, this paper accepts some network structures by adjusting the PCNN. On the premise of ensuring its biological characteristics, an improved PCNN model is proposed, which optimizes the external input of neurons, the weight of connection input L and the attenuation mode of threshold. The experimental results show that the model effectively reduces the complexity of the algorithm. The description ability of pixel spatial information is improved. (2) aiming at the fuzziness of tooth image, the number of iterations of PCNN model is difficult to determine and needs to be set manually. In this paper, based on the spatial information of image pixels, the influence of iteration times on PCNN model segmentation algorithm is analyzed, and a criterion for determining the optimal number of iterations based on minimum cross entropy is given by using the information entropy optimization criterion. The improved PCNN model segmentation algorithm is realized to segment the pulp cavity image of CBCT teeth accurately.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：D919;TP391.41

【参考文献】