【作者】 王丹；

【导师】 郭海燕；

【作者基本信息】 西南科技大学 ， 信息与通信工程， 2024， 硕士

【摘要】 随着计算机技术的发展,医疗工作者逐渐认识到常规显微镜的低效性,并将注意力转移到自动阅片系统上,这一转变促进了病理切片数字化的进程,为该领域带来广阔的发展前景。本文针对阅片效率慢和精度低等问题,重点研究了自动对焦和焦平面建模等核心算法,设计并开发了自动阅片显微成像软件系统。该系统可以实现病理切片的快速扫描,达到减少人工干预,提高工作效率的目的,具有一定的实用价值。主要工作内容如下:(1)自动对焦算法研究针对现有自动对焦技术性能不佳等情况进行改进。首先提出Tenengrad＿Variance(简称TV)评价函数,从而准确计算图像的清晰度。其次,采用中心取窗的方式减少图像计算的像素数,提高评价函数的运算速率。最后,在此基础上提出改进的混合对焦搜索算法,有效减少电机移动的次数,使其快速找到最优焦距位置完成对焦工作。本算法相较于其他对焦算法而言速度提升了1倍以上,且采集的图像峰值信噪比最大,表明该算法能够快速准确的寻找到最佳焦平面。(2)焦平面建模算法研究在病理切片数字化过程中,由于显微视野有限且单张切片往往需扫描上百个位置,才能呈现出完整的切片图像,因此面临着逐点对焦效率低下的问题。本文根据用户对图像质量的需求,设计了两种焦平面建模方法。一种是基于三角域的焦平面建模方法,该方法建模速度较快仅需0.461秒,在20x物镜下准确率为94.36%,40x物镜下准确率为93.71%,适用于能够接受少量图像离焦的用户。另一种是基于矩形结构化网格的建模方法,建模速度相对较慢需要1.490秒,但在20x物镜下准确率为99.70%,40x物镜下准确率为99.37%,适用于对图像质量要求较高的用户。(3)上位机软件平台的设计与实现本研究基于广州明美ML31-M生物显微镜和MS60-2相机作为实验平台,开发了一套自动阅片显微成像上位机软件系统。该系统集成了自动对焦算法和焦平面建模算法,具备图像采集、样本点选取、自动对焦、焦平面建模、自动扫描和图像存储等功能。其设计旨在便于用户操作使用,并为后续验证各模块算法的实施情况提供便利。更多还原

【Abstract】 With the development of computer technology,medical workers have gradually realized the inefficiency of conventional microscopes and shifted their attention to automatic film reading systems.This transformation has promoted the digitalization of pathological sections and brought broad development prospects to this field.This article focuses on the issues of slow efficiency and low accuracy in film reading,and focuses on core algorithms such as autofocus and focal plane modeling.A software system for automatic film reading microscopy imaging is designed and developed.This system can achieve rapid scanning of pathological sections,reducing manual intervention and improving work efficiency,and has certain practical value.The main job responsibilities are as follows:(1)Research on autofocus algorithmImprovements will be made to address the poor performance of existing autofocus technology.Firstly,the Tenengrad_Variance(TV)evaluation function is proposed to accurately calculate the clarity of the image.Secondly,the use of center windowing reduces the number of pixels in image calculation and improves the operation speed of the evaluation function.Finally,based on this,an improved hybrid focusing search algorithm is proposed,which effectively reduces the number of motor movements and enables it to quickly find the optimal focal length position to complete focusing work.Compared with other focusing algorithms,this algorithm has increased its speed by more than twice,and the peak signal-to-noise ratio of the captured images is the highest,indicating that this algorithm can quickly and accurately find the optimal focal plane.(2)Research on Focal Plane Modeling AlgorithmIn the process of digitizing pathological sections,due to the limited microscopic field of view and the need to scan hundreds of positions on a single slice to present a complete slice image,there is a problem of low point by point focusing efficiency.This article designs two focal plane modeling methods based on the user’s demand for image quality.One is a focus plane modeling method based on the triangular domain,which has a fast modeling speed of only 0.461 seconds and an accuracy of 94.36% under a 20 x objective and 93.71% under a 40 x objective.This method is suitable for users who can accept a small amount of image defocus.Another modeling method is based on rectangular structured grids,which has a relatively slow modeling speed and takes 1.490 seconds.However,the accuracy is 99.70% under a 20 x objective lens and 99.37% under a 40 x objective lens,making it suitable for users who require high image quality.(3)Design and Implementation of Upper Computer Software PlatformThis study developed an automatic micro imaging upper computer software system based on Guangzhou Mingmei ML31-M biological microscope and MS60-2 camera as experimental platforms.The system integrates autofocus algorithm and focal plane modeling algorithm,with functions such as image acquisition,sample point selection,autofocus,focal plane modeling,automatic scanning,and image storage.Its design aims to facilitate user operation and provide convenience for subsequent verification of the implementation of algorithms in each module.更多还原