重性抑郁障碍语音生物标志物的研究
文献类型：学位论文
| 作者 | 狄雅政 |
| 答辩日期 | 2024-06 |
| 文献子类 | 博士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 朱廷劭 |
| 关键词 | 抑郁症 遗传关联性 语音生物标志物 音高 全基因组关联分析 |
| 学位名称 | 理学博士 |
| 学位专业 | 应用心理学 |
| 中文摘要 | Major Depressive Disorder (MDD) is the leading cause of disability in the world. Yet it remains underdiagnosed. The challenge in its diagnosis primarily lies in the lack of reliable, objective biomarkers. Clinical observations have identified vocal changes as a prominent marker in patients with MDD for a long time, yet none of these observations have been effectively transformed into clinical diagnostic tools. This dissertation presents five systematic studies focused on the voice biomarkers of MDD, combining machine learning techniques and genetic analysis methods. It aimed to validate the efficacy of voice recognition technology in identifying MDD, establish a reliable association between voice features and MDD, differentiate between state and trait biomarkers for MDD, and preliminarily explore the genetic basis of voice features. Studies 1 and 2 applied the i-vectors method to characterize the global, abstract patterns of voice, aiming to validate the effectiveness of MDD detection using voice in a large population, and to preliminarily determine the genetic link between voice and MDD. By cleaning audio records from clinical interviews of a large MDD case-control cohort, we obtained voice data from 7,654 participants. Study 1 built an MDD classification model and examined the diagnostic performance of voice i-vectors and found it achieved an area under the receiver operating characteristic curve of 0.80. Combined with polygenic risk scores, the classification performance was further enhanced to 0.86, indicating the potential of voice and genetics in diagnosing MDD in large populations. Study 2 calculated the heritability of the voice i-vectors and their genetic correlation with MDD, followed by Mendelian randomization and gene enrichment analysis. The results showed significant genetic associations between the fifth coordinate of voice i-vectors and MDD. Mendelian randomization and gene enrichment analysis results suggested a causal relationship involving the axon guidance, preliminarily indicating a genetic link between voice and MDD. Study 3 further investigates the association between specific vocal features - pitch and tonal changes - and MDD. Using a two-stage meta-analysis method, the study selected 5,681 participants from 27 hospitals for association analysis and an independent sample of 1,084 participants from four hospitals for replication. This study identified and successfully replicated the association between MDD and 16 pitch features, reflecting that MDD patients had slower pitch change speeds but more frequent occurrences of extreme pitch change speeds. The analysis of risk factors in MDD patients showed that some pitch features were related to stressful life events. Based on the established associations between voice pitch and MDD in Study 3, Study 4 then differentiated between trait and state biomarkers for MDD. To discover trait biomarkers, the study utilized the genetic data from the 7,654 subjects and calculated the heritability of pitch features and their genetic correlation with MDD. To determine state biomarkers, the study used a two-stage meta-analysis method to calculate the association between pitch features and current depression symptom checklist scores, based on the 1,084 subjects in the replication cohort. The results showed that features measuring the variability and extremity of pitch change speed were heritable and had significant genetic associations with MDD, highlighting their potential use as trait biomarkers for early detection and secondary prevention of MDD. Non-heritable features like the timepoints of peak pitch change were related to the MDD state, holding the potential for clinical diagnosis of depressive episodes. Study 5 conducted a genome-wide association study (GWAS) of vocal median pitch and combined it with the results in the Icelandic population for meta-analysis. The results showed that the genetic effect of the ABCC9 locus on pitch found in the Icelandic population also existed in Chinese female samples. This effect was consistent across population, language, speaking context, and mood. The meta-GWAS further identified two novel loci, thereby deepening our understanding of the genetic control of human pitch. In summary, the findings of this dissertation not only contribute important empirical data to the field of voice biomarkers for MDD but also offer new perspectives and methodologies for further exploration of the complex relationship between voice features and MDD. The results validated the practicality of voice recognition technology in identifying MDD in large populations and revealed the genetic correlation between voice and MDD. Additionally, this research highlighted the value of specific voice features such as pitch change speed as potential biomarkers for MDD early detection and prevention, laying the groundwork for deeper research in this field. Looking forward, we hope to broaden our sample groups to include diverse populations, enhance the genetic understanding of voice biomarkers, and improve generalizability to ultimately achieve effective and timely diagnosis and prevention of MDD. |
| 英文摘要 | 重性抑郁障碍（Major Depressive Disorder, MDD，即抑郁症）是全球范围内 引发残疾的主要原因之一，其诊断挑战主要在于缺乏客观可靠的生物标志物， 导致较高的误诊率和漏诊率。长期以来，抑郁症患者展现的语音变化特征，虽 为明显的疾病迹象，却鲜少被有效转化为临床诊断工具。本论文聚焦于抑郁症 的语音生物标志物展开了五项系统性研究，结合机器学习方法和遗传学分析手 段，旨在验证语音识别抑郁症的有效性，明确语音与抑郁症之间的可靠关联， 区分抑郁症的性状与状态生物标志物，并初步探究语音的遗传基础。 研究一和研究二利用语音 i-vectors 表征语音的全局抽象特征，旨在验证语 音在大规模人群中识别抑郁症的真实效果，并初步确定语音和抑郁症的性状关 联。通过处理和筛选大型抑郁症病例-对照组队列中被试的临床访谈录音数据， 获取了 7654 名被试的有效语音样本。研究一构建了抑郁症分类模型来测试语音 i-vectors 的诊断性能，其接收者操作特征曲线下面积达到 0.80。结合多基因风险 评分，该分类效果进一步ᨀ升至 0.86，表明语音在大规模人群中具有识别抑郁 症的潜力。研究二基于被试的基因数据，计算了语音 i-vectors 的遗传度，以及 它们和抑郁症的遗传相关性，并进一步进行孟德尔随机化和基因富集分析。结 果显示，语音 i-vectors 的特定维度和抑郁症存在显著的遗传相关性。孟德尔随 机化和基因富集分析的结果表明两者具有因果关系，涉及轴突引导过程。这些 结果初步表明语音与抑郁症存在性状上的关联。 研究三进一步考察语音的局部具象特征——音高和音调变化与抑郁症的关 联。研究采用了两阶段元分析方法，从 7654 名被试中选择了来自 27 家医院的 5681 名被试进行关联分析，并在额外 4 家医院的 1084 名样本上进行复现。该研 究发现并成功复现了抑郁症和 16 个音高特征的关联，反映了抑郁症患者语音音 高变化速度较慢，但音高变化速度极端值的出现更频繁的特点。针对抑郁症患 者风险因素的关联分析显示了部分音高特征和生活压力事件次数相关。通过研 究三，我们确定了可靠的抑郁症语音生物标志物。 研究四基于研究三确定的语音生物标志物，进一步区分性状和状态生物标 志物。为了确定抑郁症的性状生物标志物，研究利用 7654 名被试的基因数据， 计算音高特征的遗传度及其和抑郁症的遗传关联性。为了确定状态生物标志物， 研究利用两阶段元分析方法，基于复现队列的 1084 名被试计算了音高特征和当 下抑郁症状分数的关联。结果显示，代表音高变化速度的变异性和极端性的特 征具有遗传性，并与抑郁症有显著遗传关联。这突显了它们作为性状生物标志 物，用于早期抑郁症检测和二级预防的潜在用途。以音调变化峰值时刻为代表的不具有遗传性的特征则与当下的抑郁状态相关，具有在临床上诊断抑郁症发 作的潜力。 研究五基于 7654 名被试对语音音高进行遗传关联分析，并结合冰岛人群的 全基因组关联结果进行元分析。结果显示，冰岛人群中发现的 ABCC9 基因上的 位点对音高的遗传效应在中国女性样本上也存在，该效应具有跨人群、语言、 情境和情绪的一致性。元分析结果进一步发现了两个全新的遗传位点，加深了 我们对于人类音高控制的遗传理解。 总结而言，本论文的研究成果不仅为抑郁症的语音生物标志物研究领域贡 献了重要的实证数据，同时也为进一步探索语音特征与抑郁症之间的复杂关系 ᨀ供了新的视角和方法。结果验证了语音识别技术在大规模人群中识别抑郁症 的实用性，并揭示了语音特征与抑郁症之间的遗传联系。此外，本研究还突出 了音调变化等具体语音特征作为早期识别和预防抑郁症的潜在性状生物标志物 的价值，为未来在这一领域的深入研究奠定了基础。展望未来，希望通过更广 泛的样本群体，加深对语音生物标志物的遗传理解，ᨀ高泛化性，最终实现对 抑郁症更及时且有效的诊断和预防。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/47977]  |
| 专题 | 心理研究所_社会与工程心理学研究室 |
| 推荐引用方式 GB/T 7714 | 狄雅政. 重性抑郁障碍语音生物标志物的研究[D]. 中国科学院心理研究所. 中国科学院大学. 2024. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。