Automated Speech Analysis: A Marker of Drug Intoxication & Treatment Outcome

自动语音分析：药物中毒的标志

• 批准号: 9232130
• 负责人: Richard W Foltin
• 金额: $ 8.1万
• 依托单位: NEW YORK STATE PSYCHIATRIC INSTITUTE DBA RESEARCH FOUNDATION FOR MENTAL HYGIENE, INC
• 依托单位国家: 美国
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-04-01 至 2018-03-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9232130
• 关键词: Acute; Behavior; Biochemical; Bypass; Cereals; Characteristics; Clinical; Clinical Research; Clinical assessments; Cocaine; Cocaine Abuse; Cocaine Users; Code; Cognitive Therapy; Complex; Computers; Computing Methodologies; Data; Data Analyses; Detection; Development; Diagnostic tests; Disease; Double-Blind Method; Drug abuse; Drug usage; Funding; Funding Mechanisms; Future; Human; Individual; Industry; Intoxication; Intravenous; Laboratory Study; Language; Lysergic Acid Diethylamide; Machine Learning; Manuals; Measures; Medicine; Mental disorders; Methods; Mind; Moods; Motivation; Natural Language Processing; Oral; Patient Self-Report; Patient risk; Patients; Pharmaceutical Preparations; Pharmacotherapy; Placebos; Prognostic Marker; Psychiatry; Psychotic Disorders; Randomized; Reporting; Research; Research Personnel; Research Proposals; Sampling; Scientist; Semantics; Source; Speech; Structure; Study Subject; Substance Use Disorder; Technology; Testing; Transcript; Treatment outcome; Work; addiction; analytical method; base; clinical practice; clinical risk; clinically relevant; cocaine use; computer science; computerized; cost; cost effective; disorder later incidence prevention; ecstasy; experience; high risk; innovation; mental state; natural language; novel; outcome prediction; predictive modeling; prognostic assays; programs; public health relevance; secondary analysis; substance abuse treatment; syntax

 DESCRIPTION (provided by applicant): A major limitation of existing assessments of clinically-relevant mental states related to drug use, abuse, and treatment is that self-report measures rely on the capacity and motivation to accurately report one's internal experiences. A potential alternative is presented by emerging computer-based natural language processing methods that can extract fine-grained semantic, structural, and syntactic features from free speech1, potentially providing a unique 'window into the mind.' These methods are widely used in industry2, yet remain largely unknown in clinical research. To begin to assess the potential of these advanced analytic methods in clinical research, we recently partnered with IBM computer science researchers to test computer-based analysis of speech semantic structure. In preliminary work, we were able to demonstrate that such methods could detect acute drug intoxication3 and accurately predicted the development of psychosis in clinical risk states4. Here, we propose to build on these highly promising initial findings, conducting three secondary data analyses to rapidly and cost-effectively advance this novel direction. Projects 1 and 2 will extend our preliminary work on speech markers of mental state changes during acute drug intoxication. In Project 1, we will assess speech semantic, structural, and syntactic features as markers of mental state changes due to MDMA (0, 0.75, 1.5 mg/kg; oral). In Project 2, we will extend these findings to another drug, assessing speech markers of intoxication with LSD (0, 70 μg; intravenous). These projects are possible because we have access to existing transcripts of free speech from within-subject, controlled laboratory studies of the effects of MDMA (N = 77) and LSD (N = 19). Potential future uses for these methods could include rapid characterization of the effects of emerging drugs and, potentially, detection of acute drug intoxication in the absence of biochemical confirmation. Project 3 will assess the use of speech analysis as a prognostic marker in substance abuse treatment. Specifically, we will use speech transcripts (N = 50) from a currently ongoing study to assess whether features extracted from baseline free speech can predict treatment outcome in cocaine users undergoing 12 weeks of CBT relapse prevention. Self-report5,6 and manual coding of speech7-9 suggest that motivation to change may be a predictor of treatment outcome for substance use disorders: we expect that the fine-grained computational methods we will employ will allow the development of more accurate predictive models. The capacity to use automated methods to detect mental states from free speech has wide ranging, potentially transformative implications for addiction medicine and psychiatry more broadly4,10. Results of the proposed secondary analyses projects will efficiently advance understanding of how automated speech analysis, a non-invasive and cost- effective assessment method, could be used in clinical practice and research about drug abuse. More broadly, results may contribute to the empirical basis for the development of automated, objective, speech- based diagnostic and prognostic tests in psychiatry.

 描述（由申请人提供）：与药物使用、滥用和治疗相关的临床相关精神状态的现有评估的一个主要限制是自我报告措施依赖于准确报告一个人的内部经历的能力和动机。是由新兴的基于计算机的自然语言处理方法提出的，这些方法可以从自由言论中提取细粒度的语义、结构和句法特征，有可能提供一个独特的“心灵之窗”。这些方法在工业中广泛使用，但仍然在很大程度上为了开始评估这些先进分析方法在临床研究中的潜力，我们最近与 IBM 计算机科学研究人员合作测试了基于计算机的语音语义结构分析，在初步工作中，我们能够证明这种方法。方法可以检测急性药物中毒3并准确预测临床风险状态下精神病的发展4。在这里，我们建议以这些非常有希望的初步发现为基础，进行三项二次数据分析，以快速且经济高效地推进这一新方向。 2 将扩展我们的前期工作急性药物中毒期间精神状态变化的言语标记 在项目 1 中，我们将评估言语语义、结构和句法特征，作为 MDMA（0、0.75、1.5 毫克/千克；口服）引起的精神状态变化的标记。 2，我们将把这些发现扩展到另一种药物，评估 LSD 中毒的言语标志物（0、70 μg；静脉注射）。这些项目是可能的，因为我们可以获得现有的自由言论记录。对 MDMA (N = 77) 和 LSD (N = 19) 的作用进行受试者内对照实验室研究，这些方法的潜在未来用途可能包括快速表征新兴药物的作用，并可能检测急性药物中毒。在没有生化确认的情况下，项目 3 将评估语音分析作为药物滥用治疗中的预后标记的用途，具体来说，我们将使用当前正在进行的研究中的语音记录（N = 50）来评估是否从基线中提取特征。演讲可以预测接受 12 周 CBT 复发预防的可卡因使用者的治疗结果 自我报告 5,6 和语音手动编码 7-9 表明，改变的动机可能是药物滥用障碍治疗结果的预测因素：我们预计，我们将采用的粒度计算方法将允许开发更准确的预测模型，使用自动化方法从自由言论中检测精神状态的能力对成瘾医学和精神病学具有更广泛的潜在变革性影响4,10。二次分析项目将有效地促进对自动语音分析（一种非侵入性且具有成本效益的评估方法）如何应用于药物滥用的临床实践和研究的理解更广泛地说，结果可能有助于为自动化、客观、自动化的发展提供经验基础。精神病学中基于语音的诊断和预后测试。