基于光学相干层析离体牙图像的去噪算法研究

Study on denoising algorithm for optical coherence tomography tooth images in vitro

目的光学相干层析成像因其高分辨率、无损等优点，适于早期龋检测；但由于系统中存在的噪声，影响其成像质量。为了重建牙齿的原貌信息，需寻找一种适于早期龋检测的光学相干层析成像实时图像去噪算法。方法比较平均曲率流滤波、非线性扩散拉普拉斯金字塔算法、非局部均值滤波3种滤波方法对光学相干层析人离体牙图像的去噪效果，从噪声抑制、边界保持、运算时间3方面分析上述3种算法的实时去噪性能及其优缺点。结果非局部均值滤波在噪声抑制和边界保持2个方面能达到很好的平衡，但实时性差；而非线性扩散拉普拉斯金字塔算法则能在滤波效果和运算效率达到较好的平衡；平均曲率流滤波次之。结论非线性扩散拉普拉斯金字塔算法较适于早期龋检测的光学相干层析成像实时图像去噪。

Objective Optical coherence tomography（OCT） is used to detect early dental caries for its high resolution and noninvasiveness. However, the imaging quality is limited due to the system noise. Therefore, in order to acquire raw dental images, it is necessary to find a real time denoising algorithm of OCT image for early caries detection. Methods A comparison of effects of three denoising algorithms, nonlinear coherent diffusion （NCD） filtering, Laplacian pyramid-based nonlinear diffusion （LPND） filtering and non-local means （NLmeans） fil- tering, for OCT tooth images was performed. Specifically, the performance of the above approaches was evaluated and compared in speckle suppression, edge preservation and time cost. Results NLmeans well balanced edges preservation and speckle suppression in spite of its high time-consuming. Meanwhile, LPND filtering provided a good compromise between image quality improvement and time cost, with the NCD filtering performed with less satisfaction. Conclusion LPND is proved to have a better potential in speckle reduction of real-time OCT tooth images for early dental caries detection.