Human-Centered Perceptual and Conceptual Classification of Biomedical Images

描述（申请人提供）：生物医学图像的数量和重要性不断增加，但缺乏有效的图像管理和理解其内容的计算解决方案。图像理解是推动这些努力的关键限制因素。主要的挑战仍然是理解人类视觉系统在生物医学成像方面的能力，以及提取和利用领域专家的隐性知识。为了应对这些挑战，我们提出了一种创新的多学科方法，将以用户为中心的设计、视觉感知和计算机成像研究方法结合起来，与领域专家进行互动，并激发和利用他们的外在和内在知识。我们将使用一种新颖的上下文设计方法来检查皮肤科图像，以发现图像的感知相关视觉内容与用户通过自然语言表达的概念理解之间的关系。对用户眼球运动和语言描述的分析，以及对领域医学本体的映射，将使我们能够将视觉数据与用户指定的语言模型集成在一起，以定义感知类别并为图像分类提供信息。这是一个基本的和具有挑战性的数据到知识的问题，尚未解决。本研究将证明通过多种感知相关模式从领域专家那里获得隐性知识的概念价值，以便整合数据和知识模型以更好地理解图像，并可能有助于在我们如何概念化和开发生物医学信息系统方面实现范式转变。

DESCRIPTION (provided by applicant): Biomedical images are ever increasing in quantity and importance yet effective computing solutions for managing images and understanding their content are lacking. Image understanding is a key limiting factor in advancing these endeavors. Major challenges remain in understanding the capabilities of the human visual system with respect to biomedical imaging and in extracting and utilizing tacit knowledge of domain experts. To meet these challenges, we propose an innovative, multidisciplinary approach which combines methods of user centered design, visual perception and computer imaging research to interact with domain experts and to elicit and use their extrinsic and intrinsic knowledge. We will use a novel contextual design approach to inspection of dermatology images to discover relationships between perceptually- relevant visual content of images and users' conceptual understanding as expressed through natural language. Analysis of users' eye movements and verbal descriptions, together with mapping to domain medical ontologies, will allow us to integrate visual data with a user-specified language model to define perceptual categories and inform image classification. This is a fundamental and challenging data to knowledge problem that has not been solved. This study will provide proof of concept of the value of eliciting tacit knowledge from domain experts through multiple perceptually relevant modes in order to integrate data and knowledge models for better image understanding and may help enact a paradigm shift in how we conceptualize and develop biomedical information systems, in general.