Development of an adaptive machine learning platform for automated analysis of biomarkers in biomedical images

摘要 研究人员和病理学家对生物医学图像的手动分析非常耗时，需要大量的时间 培训，并容易引入偏见和错误。组织样本、培养物或 标本是检测生物特性的基础，包括中央区域内的蛋白质相互作用 神经系统、精子计数、消化系统寄生虫和对病毒感染的免疫反应，如 新冠肺炎。生物标志物分析过程中的无意偏差和注意力限制可能是不良的基础 生物医学研究结果的重复性，并可能给临床诊断带来错误。这些 问题是提供最有益的循证医学的重大障碍，发展 有效的医疗，并促进公众对科学探究的信心。 将计算机视觉应用于细胞目标检测是减少人类偏见的一种有前途的方法， 主观性和错误，限制了研究的重复性，减缓了有效医学的发展 治疗。我们的图像分析软件，称为PinSqueak AITM，以及底层的人工智能(AI) 在我们的NIH SBIR第一阶段奖期间开发的技术显著增加了评分者之间和内部的评分者 多重生物标志物的组织样本分析的可靠性和分析时间的缩短。小不点人工智能是 现已集成到ImageJ/斐济(https://Pipsqueak.ai)，)，并能够返回数百个 在图像上传后300ms内向用户提供准确的蜂窝目标检测。在过去的6个月里， 小用户数量激增至每月1000多名活跃用户，这表明对计算机的需求很高 提高显微图像量化速度和精度的视觉技术。我们经过预训的ML 模型能够精确地检测多种细胞形态和目标类型 重复性大大超过了人类的分析。在这里，我们建议开发一种经过预培训的生物医学 图像分析平台，快速准确地识别不同的细胞目标，并使 一种商业形式的云计算机视觉服务，称为Sightologist.aiTM。我们的 计算机视觉人工智能即服务(AIaaS)将通过我们的 微不足道的人工智能软件和第三方产品集成。为了实现这些目标，我们将在我们的基础上 SBIR第一阶段的进展是为生物标志物检测开发ML模型，并实施云分发 向远程用户和应用程序提供我们的计算机视觉服务的方法。

ABSTRACT Manual analysis of biomedical images by researchers and pathologists is time intensive, requires intensive training, and prone to introduce bias and error. Optical analysis of targets within tissue samples, cultures, or specimens is fundamental to detecting biological properties, including protein interaction within the central nervous system, sperm counts, digestive-system parasites, and immune response to viral infections like COVID-19. Unintentional bias and attentional limitations during analysis of biomarkers can underlie poor reproducibility of findings in biomedical research and potentially introduce errors to clinical diagnostics. These problems are significant barriers to delivering the most beneficial evidence-based medicine, developing effective medical treatments, and promoting public confidence in scientific inquiry. Application of computer vision for cellular target detection is a promising approach to reducing human bias, subjectivity, and errors that limit the reproducibility of research and slow the development of effective medical treatments. Our image analysis software, called Pipsqueak AITM, and the underlying artificial intelligence (AI) technology developed during our NIH SBIR Phase I award, have significantly increased inter- and intra-rater reliability of tissue sample analysis and decreased analysis time for multiplexed biomarkers. Pipsqueak AI is available now as an integration to ImageJ/FIJI (https://Pipsqueak.ai), and is capable of returning hundreds of accurate cellular target detections to the user within 300ms of image upload. During the last 6 months, Pipsqueak usership has exploded to over 1000 active monthly users, indicating high demand for computer vision technologies that improve the speed and accuracy of micrograph quantification. Our pre-trained ML models are capable of detecting multiple cellular morphologies and target types with precision and reproducibility that greatly exceed human analysis. Here, we propose to develop a pre-trained biomedical image analysis platform that rapidly and accurately identifies diverse cellular targets, and make this technology commercially available as a cloud computer vision service, called Sightologist.aiTM. Our computer vision AI-as-a-service (AIaaS) will be made available to research and clinical end-users through our Pipsqueak AI software and through 3rd party product integrations. To achieve these goals, we will build on our SBIR Phase I progress that developed ML models for biomarker detection, and implement cloud distribution methods to deliver our computer vision service to remote users and applications.