Intelligent and Automatic Image Segmentation Software for High ThroughputAnalysi

描述(由申请人提供)：众所周知，衰老和许多慢性疾病，如癌症和心力衰竭，与人类骨骼肌质量和力量的显著下降有关。肌肉生物界一致认为，肌肉纤维的重要形态特征，如纤维面积、肌核的数量和位置、细胞浸润和纤维化是决定肌肉健康和功能的关键因素。然而，目前，从标准的组织学和免疫组织学技术对肌肉特征进行量化仍然是一个手动或充其量是半自动的过程。这一过程是劳动密集型的，可能容易出错，导致观察者之间的高度可变性。另一方面，当肌肉特征通过计算机辅助图像分析计算时，数据采集时间减少，客观性显著提高。该项目第一阶段的目标是建立一个全自动、智能和高通量的图像采集和分析软件，用于数字化肌肉横断面的定量肌肉形态分析。我们建议利用机器学习和生物医学图像分析方面的最新技术进步。这包括新开发的可变形模型和基于均值漂移的种子检测算法，以提高分割精度；基于非对称在线Boosting的机器学习算法，允许软件从错误中学习并自适应调整其分割策略；以及使用图形处理单元(GPU)来处理超大规模图像(如整个幻灯片扫描样本)的计算瓶颈的数据并行化方案。我们相信，该软件配备了最先进的技术创新，将对包括基础科学家、临床科学家和制药行业在内的骨骼肌研究社区具有商业吸引力。其具体目标是：1)开发、实施和验证骨骼肌组织自动生物图像分析软件包；2)开发新型在线更新的智能人工智能单元，使软件能够从错误中学习；3)构建新型高性能计算单元，实现快速、高通量的自动图像分析，能够处理整个切片扫描的肌肉标本。提出的分析方法将为骨骼肌形态特征提供更一致、准确和客观的量化，单机版数据分析时间将减少100倍以上，并行版本数据分析时间将减少2000倍以上。细胞信息学LLC第二阶段的长期目标是应用该软件分析人类肌肉活组织样本的组织学/病理学，并将该软件扩展到其他生物组织，如脂肪组织。

DESCRIPTION (provided by applicant): It is well established that aging and many chronic diseases, such as cancer and heart failure, are associated with significant losses in skeletal muscle mass and strength in humans. There is agreement across the muscle biology community that important morphological characteristics of muscle fibers, such as fiber area, the number and position of myonuclei, cellular infiltration and fibrosis are critical factors that determine the health and function of the muscle. However, at this time, quantification of muscle characteristics from standard histological and immunohistological techniques is still a manual or, at best, a semi-automatic process. This process is labor intensive and can be prone to errors, leading to high inter-observer variability. On the other hand, when muscle characteristics are calculated by computer-aided image analysis, data acquisition times decrease and objectivity improves significantly. The objective of this Phase I STTR project is to build a fully automatic, intelligent, and high throughput image acquisition and analysis software for quantitative muscle morphological analysis on digitized muscle cross-sections. We propose to utilize the most recent technical advances in machine learning and biomedical image analysis. This includes a newly developed deformable model and mean-shift based seed detection algorithm for better segmentation accuracy; an asymmetric online boosting based machine learning algorithm which allows the software to learn from errors and adjust its segmentation strategies adaptively; and a data parallelization schema using the graphic processing unit (GPU) to handle the computational bottleneck for extremely large scale image, such as whole slide scanned specimens. We believe that this software, equipped with the most advanced technical innovations, will be commercially attractive for the skeletal muscle research community including basic scientists, clinician scientists, and the pharmaceutical industry. The specific aim are: 1) Develop, implement, and validate an automatic biological image analysis software package for skeletal muscle tissue; 2) Develop a novel online updated intelligent artificial intelligence unit to enable the software to learn from errors; 3) Build a novel high performance computing unit to enable fast and high throughput automatic image analysis, which is capable of processing whole slide scanned muscle specimens. The analysis approach proposed will provide more consistent, accurate, and objective quantification of skeletal muscle morphological properties and the time for data analysis will be reduced by over a factor of 100 for standalone version and 2000 for parallel version. The long-term goal of Cytoinformatics, LLC for the Phase II stage is to apply the software to analyze histology/pathology from human muscle biopsy samples and extension of the software to other biological tissues, such as adipose tissue.