Advancing Computational Linguistic Biomarkers of Disorganized Speech in Psychosis

项目摘要 精神错乱具有重要的临床意义，但研究还不够深入。有几条证据 提示组织紊乱反映了更高的遗传负荷和更糟糕的结果，并且对治疗敏感 反应和复发。我们将使用计算语言学来衡量一个敏感的， 客观、高效、可重复、可重复的方法。演讲将以开放式和结构化的方式引出 来自270名精神分裂症谱系障碍和带有精神病特征的情绪障碍患者的任务， 在整个样本中生成了大约30,000个句子。调查结果将在现有的独立 数据集。我们将在两个样本中测量精神病症状、功能和认知。语无伦次和 低效将被贴上个别句子的标签，并对整个参与者进行评级。我们的具体目标是 具体如下：(1)开发深度学习方法，对句子级错乱进行分类；(2)整合 预测参与者级组织混乱的计算特征；(3)预测关键参与者特征 使用语言特征。综合培训计划将通过这些研究结合实践经验 旨在通过指导、课程安排、自学、研讨会和会议来实现以下培训 目标：(1)精通计算语言学和机器学习方法；(2)具备验证方面的专业知识 与临床相关的生物标志物，以及(3)作为内科科学家的发展。这项工作提供了基础 需要将尖端计算方法发展成组织混乱和关键精神病的生物标记物 结果。我们为利用这些特征作为治疗早期标志的未来研究奠定了基础 反应和复发，并使用这些特征将行为表型与潜在的生物学联系起来。 拟议的项目建立在我现有的专业知识基础上，以提高我在以下方面的技术熟练程度和专业知识 精神病生物标记物研究我需要作为一名独立医生领导这一领域的新发现- 调查员。

Project Summary Disorganization in psychosis has important clinical implications but is under-studied. Several lines of evidence suggest that disorganization reflects higher genetic loading and worse outcomes, and is sensitive to treatment response and relapse. We will use computational linguistics to measure disorganization in a sensitive, objective, efficient, reproducible, and repeatable way. Speech will be elicited with open-ended and structured tasks from 270 people with schizophrenia spectrum disorders and mood disorders with psychotic features, generating ~30,000 sentences across the sample. Findings will be validated in an existing independent dataset. We will measure psychosis symptoms, functioning, and cognition in both samples. Incoherence and inefficiency will be labeled for individual sentences and rated for the overall participant. Our Specific Aims are as follows: (1) Develop deep-learning methods to classify sentence-level disorganization; (2) Integrate across computational features to predict participant-level disorganization; (3) Predict key participant characteristics using linguistic features. An integrated training plan will combine hands-on experience through these research aims with mentorship, coursework, self-study, seminars, and conferences to achieve the following Training Goals: (1) Proficiency in computational linguistics and machine learning methods, (2) Expertise in validating clinically-relevant biomarkers, and (3) Development as a physician-scientist. This work provides the foundation needed to develop cutting-edge computational methods into biomarkers of disorganization and key psychosis outcomes. We lay the groundwork for future studies that leverage these features as early markers of treatment response and relapse, and that use these features to connect behavioral phenotypes with underlying biology. The proposed project builds on my existing expertise to develop the technical proficiency and expertise in psychosis biomarker research I need to lead new discoveries in this area as an independent physician- investigator.