Digital image analysis for quantitative and qualitative assessment of pig islets

描述（由申请人提供）：埃德蒙顿小组证明，人胰岛移植可成功用于治疗患有难治性低血糖的成人1型糖尿病患者，这导致临床试验和进一步研究的资金增加，以通过使用猪胰岛代替人胰岛来扩展该治疗的范围。在改善免疫抑制治疗方案方面已经取得了重大进展，使得用人胰岛移植治疗成人糖尿病患者获得的结果与胰腺移植后获得的结果相似。将这种疗法从临床研究转移到常规临床实践的主要障碍是提高从人或猪胰腺回收的胰岛的产量和质量。目前，还没有标准化的方法可以准确地评估胰岛移植过程中使用的胰岛的数量或质量，从而可以客观地评估实验室之间的结果。这项资助的重点是开发一个强大的胰岛图像分析软件，以客观地分析从胰腺中回收的猪胰岛的数量和质量。该项目的两个主要目标是首先开发一种改进的图像分析软件程序，该软件程序将提供细胞制剂中猪胰岛数量和质量的标准化测量。第二，通过将每个猪胰岛的图像特征与人工类别相关联来增强软件程序的能力。将手工挑选相似大小的猪胰岛，并根据双硫腙染色的形状、边界、完整性或均匀性分为三类。第一个软件增强将在图像中找到那些可用于区分不同类别胰岛的特征。第二次增强将评估使用机器学习方法将这些特征与从图像中恢复的数据以及用于表征猪胰岛制剂的其他离散或连续变量相关联的可行性。如果成功，使用快速和客观的图像分析方法的能力将改善实验室内和实验室之间胰岛数量和质量的评估;将图像特征与移植成功（通过移植物存活率和胰岛素独立性测量）相关联;并改善胰岛分离方法，以获得通过回顾性分析确定的有利的胰岛图像评分。一家专注于通过专注于组织解离来提高胰岛产量的商业公司的能力与一家领先的学术实验室合作，该实验室在从显微图像开发软件算法方面具有复杂的专业知识，为需要解决的困难医疗提供了一种新的方法，以实现胰岛移植治疗成人1型糖尿病患者的全部潜力。 公共卫生相关性：一种客观、可靠和准确的方法来评估胰岛的数量和质量，对于胰岛移植作为1型糖尿病治疗的标准化和随后的成功至关重要。使用具有校准目镜分划板的光学显微镜确定胰岛产量的常规手动方法是主观的、耗时的，并且由于在将胰岛数量转换为胰岛当量时的采样误差和错误假设而经常高估胰岛质量。这项研究将利用数字图像分析的最新进展，包括机器学习和模式识别，开发一种软件算法，用于快速表征用于移植程序的胰岛。

DESCRIPTION (provided by applicant): The demonstration by the Edmonton group that human islet transplantation can be successfully used to manage adult type 1 diabetes patients with refractory hypoglycemia has led to increased funding of clinical trials and further research to extend the scope of this therapy by using porcine islets in place of human islets. Significant advances have been made in improving immunosuppression treatment regimens so that results obtained from treating adult diabetic patients with human islet transplants are similar to those obtained after pancreas transplantation. The major hurdle to move this therapy from clinical research to routine clinical practice is to improve the yield and quality of islets recovered from human or porcine pancreas. Presently, there are no standardized methods that can accurately assess the number or quality of islets that are used in the islet transplantation procedures so that results between laboratories can be objectively evaluated. This grant is focused on developing a robust, islet image analysis software to objectively analyze the number and quality of porcine islets recovered from the pancreas. The two major aims of the project are first to develop an improved image analysis software program that will provide a standardized measurement of the number and mass of porcine islets in a cell preparation. And second, enhance the capabilities of the software program by correlating the image signatures of each porcine islet to an artificial category. Porcine islets of similar size will be handpicked and sorted into three categories based on the shape, border, integrity, or uniformity of dithizone staining. The first software enhancement will find those features in the images that can be used to distinguish the different categories of islets. The second enhancement will assess the feasibility of using machine learning methods to correlate these features with data recovered from the images but also other discrete or continuous variables that are used to characterize the porcine islet preparations. If successful, the ability to use a rapid and objective image analysis methodology will improve the assessment of the number and quality of islets within and between laboratories; correlate image features with success of transplantation as measured by graft survival and insulin independence; and improve the islet isolation methods to achieve favorable islet image scores that are determined by retrospective analysis. The ability of a commercial firm focused on improving islet yields by focusing on tissue dissociation with a leading academic laboratory that has sophisticated expertise in developing software algorithms from microscopic images provides a fresh approach to a difficult medical that needs to be resolved to realize the full potential of islet transplantation to treat adult type 1 diabetic patients. PUBLIC HEALTH RELEVANCE: An objective, reliable and accurate method for the assessment of islet quantity and quality is paramount to the standardization and subsequent success of islet transplantation as a treatment for type 1 diabetes. Conventional manual methods for determining islet yields using an optical microscope with a calibrated eyepiece reticule are subjective, time consuming and often overestimate islet mass due to sampling errors and erroneous assumptions in the conversion of islet numbers to islet equivalents. The research proposed will utilize recent advances in digital image analysis, including machine learning and pattern recognition, to develop a software algorithm for the rapid characterization of islets destined for transplantation procedures.