SBIR Phase II: Augmenting Virtual Healthcare with Voice Biomarkers

小型企业创新研究(SBIR)第二阶段项目的更广泛影响是，通过利用机器学习和人工智能来简化临床决策支持，在美国发展智能、强大的医疗基础设施。语音生物标记器检测各种健康状况、情绪和疾病，并为实时分诊提供独特的无缝反馈。将语音语调转换为语音生物标志物将使疾病预测和监测成为可能。建议的语音生物标记器技术可能是一种可扩展的行为健康筛查，在所有虚拟护理就诊中提供公平的护理，缓解慢性疾病3T美元中80%的复杂且昂贵(2-3倍)的抑郁和焦虑并存。这个小型企业创新研究(SBIR)第二阶段项目致力于在初级保健中提供可扩展的精神健康筛查。研究目标是从全球语音生物标记物数据结合唯一的纵向元数据了解慢性健康状况的潜在行为健康触发因素。这项拟议研究中的主要技术挑战包括：(1)收集关于环境和生理变量的足够多样化的元数据标签；(2)通过主成分分析基于性别、年龄和其他具有高方差的特征训练不同的模型；(3)在设计、验证和部署阶段识别并最小化稀疏人群的偏差；以及(4)在各种呼叫中心、远程医疗平台、远程患者监控和护理管理平台模式中改进当前语音生物标记物诊断的敏感度、特异度和可诊断性。医疗保健中的基础设施部署高度复杂，需要针对特定健康人群调整多个模型，并深入了解经典和深度学习技术，以提高在看不见的人群中的准确性和泛化能力。在解决这一系列极具挑战性的机器学习任务方面取得的预期技术成果，对于实时分诊和大规模获得可靠的精神卫生保健具有深远的意义。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase II project is to develop smart, robust healthcare infrastructure in the U.S. by leveraging machine learning and artificial intelligence to streamline clinical decision support. Voice biomarkers detect a variety of health conditions, emotions, and diseases and provide a unique, seamless feedback for real-time triage. Transforming voice intonations into voice biomarkers would allow disease prediction and monitoring. The proposed voice biomarker technology is potentially a scalable behavioral health screener to provide equitable care in all virtual care visits, mitigating the complex and costly (2-3X) comorbidities of depression and anxiety in 80% of $3T in chronic conditions. This Small Business Innovation Research (SBIR) Phase II project is dedicated to providing scalable mental health screening in primary care. The research objectives are to understand the underlying behavioral health triggers for chronic health conditions from global voice biomarker data combined with unique, longitudinal metadata. The major technical challenges in this proposed research include (1) collecting sufficiently diverse metadata labels on environmental and physiological variables,(2) training distinct models based on gender, age, and other features that have high variance through principal component analysis, (3) identifying and minimizing bias for sparse populations in design, validation, and deployment phases, and (4) improving the current voice biomarker diagnostic on dimensions of sensitivity, specificity, and diagnosability in various call center, telehealth platform, remote patient monitoring, and care management platform modalities. The highly complex deployments across infrastructure in healthcare require multiple models tuned for specific health populations and a deep understanding of classical and deep learning techniques for improving both accuracy and generalizability across unseen populations. The anticipated technical results in solving this series of highly challenging machine learning tasks is profound for real-time triage and access to reliable mental healthcare at scale.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.