数字病理学图像细胞核检测、分割及分类问题的关键技术研究

项目名称： 数字病理学图像细胞核检测、分割及分类问题的关键技术研究

项目编号： No.61472411

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 辜嘉

作者单位： 中国科学院深圳先进技术研究院

项目金额： 80万元

中文摘要： 病理学分析是肿瘤诊断和分期的金标准，近年来，基于显微镜观察的传统病理学已逐步发展为基于数字成像和图像分析技术为的数字病理学，因为后者不但能进行定性分析，还能进行定量分析。定量分析的核心技术问题是图像中细胞核的检测、分割和分类。目前该领域已有不少研究成果，却鲜见其真正应用在临床实践中，究其原因，主要有三：1，缺乏统一的benchmark方法评价算法的可用性；2，未考虑到交互性设计，所以算法的鲁棒性和准确性不高；3，计算速度太慢。本研究拟解决以上这三个问题。首先，我们将设计完善的benchmark系统对不同算法进行客观评价；其次，我们将设计交互式算法接口，使得每一种算法可以调用我们的接口进行高效交互；再者，我们将设计完善的GPU计算框架，来支持算法设计者可以简单地将算法移植到GPU，从而获得数倍以上的性能提升。本研究的成果有助于实现本领域内算法的客观评测，并帮助提高这些算法和方案的临床可用性。

中文关键词： 数字病理学；基准测试；GPU并行；图像分割

英文摘要： Pathological analysis remains the golden standard of tumor diagnosis and staging. Recently conventional microscopy based pathology has been gradually developed to digital pathological, which is based on digital imaging and image analysis techniques. Digital pathology can support quantitative analysis, for which the key technique is the detection, segmentation and recognition of nucleus. In the past decade, lots of algorithms have been proposed, but we still can hardly find their applications in clinical practice. This is caused by three reasons: 1) it is lacking of a benchmark tool or framework to assert the performance of different algorithms; 2) these algorithms are lacking of interactive design to improve their stability and accuracy; 3) these algorithms are too slow for real use. We plan to address these three problems by conducting this research. First we will design a comprehensive benchmark framework and software to score different algorithms in different cases; second we will design an efficient interactive solution to support pre and post processing of different algorithm phases; finally we will design a GPU computing framework to help the algorithm developers to easily immigrant their algorithms from CPU to GPU, and then obtain performance improvement. The output of this research project will drive objective assessment of the algorithms and solutions in this domain, and help to improve their clinical utility.

英文关键词： Digital Pathology;Benchmark;GPU Computing;Image Segmentation