基因芯片图像处理的图式主动分析研究

Image processing is a key aspect of cDNA microarrays analysis and the aim of the whole experimental procedure is to obtain meaningful biological conclusions, which depends on the accuracy of this stage. To enhance the accuracy, the learning based methods arise continuously. Unfortunately, owing to the complicated hybridisation procedure of microarrays will result in dynamic changing of microarray image quality, the results of existing methods can't meet the need of overall accuracy and batch processing. .Inspired by the "perception-planning-motion" of schema theory in recognition sciences, to establish an active image analyzing method with the feature of self-organization, self-learning and self-adaption, this project analyzes the potential factors causing uncertainty of image processing results from two aspects: data and methods. To be specific, this study includes: firstly, propose feature extraction methods for schema perception by digging and analyzing the intrinsic reasons of uncertainty, and build up the quality evaluation mechanism by taking into account the image feature and the hierarchy structure of image processing. Furthermore, realize the off line parameter identification by using the Pareto optimization method and evolution algorithm, and the on-line correction based on data driven. Secondly, explore the new and novel image processing algorithm by target significant analyzing based on visual attention mechanism model, energy function and the variational level set. Finally, build up the system coordination based on schema and realize the associative mapping of character-method. In addition, develop a software tool and prove its validity by using comparison experiments. .This proposal aims to build up a method foundation for active image analyzing and processing, and provide research idea and method references for similar biological and medical image processing..

图像处理作为基因芯片分析的关键步骤，其结果准确性决定了后续生物学结论的可信度。带有学习功能的新方法不断涌现以提高准确率，而基因芯片的复杂制备过程造成图像质量动态变化，使得现有方法处理结果无法满足其全局准确和批量处理要求。本项目从数据和方法层面分析结果不确定性的可能来源，借鉴认知科学图式理论的“感知-关联-运动”，构建自组织、自学习和自适应的图式主动分析体系，具体包括：挖掘分析产生不确定的内在原因，结合图像自身特点及处理过程的分级特性，研究感知特征提取与分类方法，建立结果评价机制，结合帕累托最优采用进化算法实现离线参数辨识，并基于数据驱动进行在线修正；结合视觉注意机制模型、能量泛函及变分水平集等技术进行目标显著性分析，探索基因芯片图像处理新算法；基于图式建立系统协调，实现特征-方法关联映射，并构建工作流平台。本项目旨在为主动图像分析处理建立方法基础，为生物医学图像处理提供研究思路和方法参考。

基因芯片图像处理对提取基因表达水平、解读成千上万个基因信息、探索致病基因、建立可信疾病预测具有重要意义。而基因芯片的复杂制备过程导致图像质量多变（低对比度、强表达块状噪声、基因点形态各异、基因点分布疏密不一），使得图像处理准确率不高。本项目借鉴智能感知，基于图式理论的动觉、表象与思维，围绕基因芯片图像处理的不确定性开展研究，研究了图像质量感知特征提取、图像处理方法结构体系及软件平台设计。首先建立了基因芯片图像质量主动增强体系，引入四阶矩建立图像对比度自动感知图式，结合数据驱动学习与模式分类建立关联图式，设计了基于背景灰度估计的对比度增强运动图式。其次，针对基因点分割建立了多特征融合感知图式，结合主成分分析设计了关联图式，基于无监督聚类实现基因点分割运动图式。最后，基于深度特征挖掘与学习实现了基因点图像分割的评价与自适应调整。本项目所建立的基因芯片图像主动对比度增强方法无需人工干预，具有广泛适用性，可推广应用到其他类型图像上。本项目探索的结合无监督聚类与有监督深度学习的基因点分割方法明显优于现有方法。这些研究为获取有效基因表达谱数据奠定了基础，有利于基因芯片的生物学与病理分析推广应用。