Novel Methods for Automated Key Image Selection

描述(由申请人提供)：近年来，由于成像技术的快速发展，关于人类行为和大脑的重要新知识已经曝光。随着成像在神经科学中的作用变得越来越重要，管理和检索图像的有效方法将变得更加关键；如果没有这样的进展，进一步的进展将受到阻碍。该建议的目标是自动汇总大型成像集。图像摘要提供了一种压缩成像研究的方法，只选择相关的图像切片来客观记录患者的病情；因此，它的应用包括多媒体电子病历、远程医疗和教学文件。在第一阶段，开发的重点是一种可定制的脑图谱，用于登记患者的成像研究，以便选择关键图像。这一阶段涉及从“正常”研究和只有细微形态变化的研究中选择图像，这是大多数精神障碍患者的典型特征。除了用于将图谱映射到患者的原始研究的配准方法之外，还开发了用于定制图谱以获得成像研究获取参数的自动技术。在这项初步工作的基础上，第二阶段扩大到包括从通过主成分分析显示大体形态变化的“异常”研究中选择图像，进一步为不同年龄组(例如儿科)定制地图集，并纳入医疗报告的结构化数据输入(SDE)和自然语言处理(NLP)，以帮助指导自动选择关键图像。由此产生的产品将是一个完全自动化的软件系统，可以从任何成像研究中选择相关图像。第一阶段的初步评价将与人类专家相比，审查对比可定制地图集和摘要/相关切片选择的表现。

DESCRIPTION (provided by the applicant): Significant new knowledge about human behavior and the brain has come to light in recent years, due in part to rapid technical developments in imaging. As the role of imaging becomes increasingly important in neurosciences, effective methods for managing and retrieving images will become even more critical; without such advances, further progress will be hindered. The goal of this proposal is the automated summarization of large imaging sets. Image summarization proffers a method to compress imaging studies by selecting only pertinent image slices that objectively document a patient's condition; as such, its applications include multimedia electronic medical records, telemedicine, and teaching files. In Phase I, development is focused on a customizable brain atlas used for registering patient imaging studies in order to select key images. This phase addresses selection of images from "normal" studies and studies with only subtle morphological changes, as typical of most patients with psychiatric disorders. Automatic techniques for customizing the atlas to imaging study acquisition parameters are developed, in addition to registration methods for mapping the atlas to the patient's original study. Building from this initial work, Phase II expands to encompass selection of images from "abnormal" studies that exhibit gross morphological changes through principle component analysis, further customization of the atlas for different age groups (e.g., pediatric), and incorporation of structured data entry (SDE) and natural language processing (NLP) of medical reports to help guide automatic selection of key images. The resultant product will be a fully automated software system that can select relevant images from any imaging study. Initial evaluation in Phase I will examine the performance of the contrast customizable atlas and summarization/relevant slice selection, as compared to human experts.