用于查和监测双极性抑郁症的自动语音分析:机器学习模型开发和解释研究
Sooyeon Min1, Tae-Sung Yeum2, Daun Shin3
1Department of Neuropsychiatry, Seoul National University Hospital, Seoul, Republic of Korea.
JMIR medical informatics
|December 4, 2025
概括
语音分析可以检测双极性抑郁症和复发. 多模式语音分析为心理健康监测和护理提供了一种可扩展的方法.
科学领域:
- 精神病学和心理健康 精神病学和心理健康
- 计算语言学 计算语言学
- 数字健康数字健康
背景情况:
- 双极性障碍 (BD) 抑郁症发作会导致显著的功能障碍和降低生活质量.
- 早期和客观地发现双极性抑郁症对于及时干预至关重要.
- 多模式语音分析显示了在BD抑郁症中识别精神运动,认知和情感变化的潜力.
研究的目的:
- 开发用于选双极性抑郁症和监测纵向变化的分类器.
- 使用语音标记检测双相情绪障碍患者的抑郁复发.
- 比较不同语音模式的预测性能.
主要方法:
- 在1年内从92名BD患者收集了304个语音录音.
- 从语音中提取声学特征 (openSMILE) 和语言特征 (LIWC).
- 开发了使用极端梯度增强和光梯度增强的个人间和内部分类器,通过引导式交叉验证进行验证.
主要成果:
- 语音分析发现,抑郁症患者的能量调节减少,单调和消极情绪词语的使用增加.
- 综合声学和语言分类器检测到中度至重度抑郁症 (AUC=0.76).
- 个人内部分类器检测到抑郁症复发 (AUC=0.70).
结论:
- 语音标记可以有效地检测和监测双极性抑郁症和复发.
- 对转录和翻译语音的心理语言学分析是跨语言的可行.
- 自动语音分析为心理健康监测提供了一个可扩展的数字健康解决方案.
相关概念视频
Mania and Antimanic Drugs: Overview
530
Mania, a psychological condition characterized by elevated mood, increased energy, and reduced sleep need, is part of the bipolar disorder cycle. The exact cause of mania isn't entirely known, but it is thought to be a combination of genetic, environmental, and neurological factors. Bipolar disorder involves alternating manic and depressive episodes. Mood stabilizers like lithium, antipsychotics, and anticonvulsants help manage these episodes. Lithium carbonate is particularly effective as...
530
Bipolar Disorder
602
Bipolar disorder is a chronic mental health condition marked by significant mood fluctuations, including episodes of mania and depression. Elevated energy levels, heightened mood or irritability, impulsive behavior, reduced sleep needs, rapid speech, racing thoughts, inflated self-esteem, and distractibility characterize mania. Individuals with bipolar disorder often alternate between depressive and manic states, with periods of emotional stability lasting an average of six months to a year.
602


