EAGER: Feasibility of Using Speech as Biomarker for Concussions

该项目研究了使用语音作为脑震荡新生物标志物的可行性，基于先前研究的见解，这些研究表明大脑功能受损与语音之间存在联系。轻度创伤性脑损伤（mTBI），如脑震荡，可以说是当今体育运动中最紧迫的问题之一，仅在青少年体育运动中，每年估计就有200万至400万例。对脑损伤患者的健康和福祉的潜在短期和长期影响是广泛的。未经治疗的脑震荡会导致慢性创伤性脑病、阿尔茨海默病发病率升高以及痴呆症的发病率更高和更年轻。因此，必须准确检测和适当治疗mTBI，特别是在青少年中，他们的大脑仍在发育，更容易受到长期损害。这项研究推进了对脑损伤与语音声学特征之间关系的理解，并为新颖准确的脑震荡筛查工具奠定了基础。基于来自300多名拳击手（其中25名是脑震荡）的语音收集的初步结果是有希望的，并表明语音可能有潜力作为一种新的，易于捕获的mTBI生物标志物。使用智能手机和平板电脑等移动的技术，青年运动员的简短演讲录音正在被捕捉。在语音识别技术的帮助下，每个单词被隔离，并为每个单词中的每个音素提取声学特征。被分析的特征包括语音的时间特性（例如，每个单词或音素的语音持续时间或发音率）和频谱特性（例如，音调、共振峰频率和梅尔频率倒频谱系数）。使用最先进的统计分析和机器学习技术，然后分析这些特征作为轻度脑损伤标志物的潜力。

This project studies the feasibility of using speech as a novel biomarker for concussions, based on insights obtained from prior research that have shown links between impaired brain functioning and speech. Mild traumatic brain injuries (mTBI) such as concussions are arguably one of the most pressing concerns in sports today, with an estimated 2-4 million cases every year in youth sports alone. The potential short- and long-term impacts on the health and well-being of individuals with brain injuries are extensive. Untreated concussions can lead to diseases such as chronic traumatic encephalopathy, an elevated incidence of Alzheimer's disease, and dementia developing at a higher rate and a younger age. It is therefore imperative to accurately detect and appropriately treat mTBI, especially among adolescents, whose brains are still developing and more prone to long-term damage. This research advances the understanding of the relationship between brain injuries and acoustic features in speech and lays the foundation for novel and accurate concussion screening tools. Initial results based on voice collections from more than 300 boxers (25 of which were concussed) have been promising and indicate that speech may have the potential to serve as a new, easy-to-capture biomarker for mTBI. Using mobile technologies such as smartphones and tablets, brief speech recordings from youth athletes are being captured. With the help of speech recognition techniques, each word is isolated and acoustic features are extracted for each phoneme in each word. Features that are being analyzed include temporal characteristics of speech (e.g., speech duration per word or phoneme or articulation rate) and frequency spectrum characteristics (e.g., pitch, formant frequencies, and mel-frequency cepstrum coefficients). Using state-of-the-art statistical analysis and machine learning techniques, these features are then analyzed for their potential as marker for mild forms of brain injuries.