CAREER: Advanced data analytics for early detection of Alzheimer's disease using wearables and smartphone

阿尔茨海默病（AD）每年给美国经济带来至少2500亿美元的负担。预计到2050年，其流行率将增加两倍，除非进一步的发现有助于预防这种疾病。由于AD筛查的挑战，它通常无法诊断，直到病例进展和足够的神经元损伤已经发生，疾病的逆转是不可能的。迫切需要的是一种认知筛查工具，可以客观地检测高危人群并监测疾病进展，而无需专业人员或医疗设备。在可穿戴传感器的快速发展及其日益增长的社会可接受性的推动下，该CAREER计划将开发早期检测AD所需的新型数据分析方法。这种工具可以启动进一步的生物标志物测试和早期干预，以便在为时已晚之前减轻神经元损伤。此外，一个综合的教育和推广计划旨在促进跨学科研究培训，并增加代表性不足的学生在STEM学科的参与。该项目将通过学校示范、校园实验室研究和暑期体验，创建针对初中和高中学生的学习模块，重点是让女学生在早期参与。它将把研究成果纳入工程课程，重点是神经退行性疾病，以激发女性和少数民族学生对工程教育的兴趣。从该项目中获得的知识将通过研究出版物，教程和研讨会以及更广泛的外展活动传播到该地区的公共图书馆，科学博物馆和利益相关者。该项目是一个重大的偏离，目前的努力，以解决目前的技术障碍，在个性化和纵向监测。目前的数据分析方法不考虑个人在其方法中的变异性。它们使用来自一组特征良好的受试者的数据进行训练，然后应用于新的受试者，而不管差异如何。因此，他们没有考虑受试者之间模式的变化，这对于检测由于疾病与受试者内差异引起的异常模式至关重要。另外，现有方法中没有一个考虑可能随时间发生的受试者内变化。这种限制对于认知障碍严重程度的纵向监测尤其重要，因为这些方法无法区分纵向疾病相关变化与衰老引起的变化。在数据分析和临床实践中解决这些挑战的计划方法如下。首先，该项目通过开发创新的个性化深度学习方法，在认知衰退的初步评估期间显著提高了对高危个体的早期检测。这些新方法将基于编码-解码器架构，以探索与认知下降相关的特征的原始步态数据，同时将开发一种新的域自适应技术，以根据个人的可变性定制架构。其次，它通过开发新的自适应深度学习方法来纵向监测认知下降，从而显着提高了疾病进展率的检测。该方法将基于具有异构数据融合能力的深度学习方法，以实现自由行走步态的分析，沿着语音数据。将开发一种基于强化学习技术的新型无监督域自适应，以根据随时间推移的受试者内变化来定制模型。该研究的关键变革方面是基于现有传感器数据的数据分析的发展，但它提供了个性化的模型，显示随着时间的推移而发生的变化。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Alzheimer’s disease (AD) burdens the United States economy by at least $250 billion annually. Its prevalence is predicted to triple by 2050 unless further discoveries facilitate the prevention of the disease. Because of challenges in screening for AD, it often remains undiagnosed until cases advance and sufficient neuronal injury has occurred that reversal of the disease is unlikely. What is desperately needed is a cognitive screening tool that can objectively detect at-risk individuals and monitor the disease progression with no specialized staff or medical equipment. Driven by rapid advancements in wearable sensors and their growing social acceptability, this CAREER program will develop novel data analysis approaches required for the early detection of AD. Such a tool could initiate further biomarker testing and early interventions in order to alleviate neuronal damage before it becomes too late. In addition, an integrated educational and outreach program is designed to both foster interdisciplinary research training and to increase participation of underrepresented students in STEM disciplines. This project will create learning modules to target middle and high school students through school demonstrations, on-campus laboratory research, and summer experiences, with a focus on engaging female students at an early stage. It will integrate research outcomes in engineering curriculum with a focus on neurodegenerative diseases in order to ignite interest in engineering education among female and minority students. Knowledge gained from this project will be disseminated via research publications, tutorials, and workshops, as well as broader outreach activities to the area public library, science museum, and stakeholders.This project is a major departure from current efforts to address current technical obstacles in individualized and longitudinal monitoring. Current data analysis approaches do not consider individuals' variability in their methods. They are trained using data from a well-characterized group of subjects and then applied to a new subject, regardless of differences. As a result, they do not consider the variation of patterns between subjects, which is critical for the detection of abnormal patterns due to the disease versus the differences within a subject. Additionally, none of the existing approaches consider intra-subject variations that may occur over time. This limitation is particularly critical for longitudinal monitoring of the cognitive impairment severity because these approaches are unable to distinguish longitudinal disease-related changes from changes due to aging. The planned approach to address these challenges in both data analytics and clinical practice is as follows. First, the project significantly improves the early detection of at-risk individuals during initial assessments of cognitive decline by developing innovative individualized deep learning approaches. These novel approaches will be based on an encode-decoder architecture to explore raw gait data for features related to cognitive decline, while a novel domain adaptation technique will be developed to customize the architecture according to an individual's variability. Second, it significantly improves the detection of the disease progression rate by developing novel adaptive deep learning approaches for longitudinal monitoring of cognitive decline. The approach will be based on deep learning methods with heterogeneous data fusion ability to enable the analysis of free-walking gait, along with speech data. A novel unsupervised domain adaptation based on reinforcement learning techniques will be developed to customize the model according to the intra-subject variations over time. The key transformative aspect of the proposed research is the development of data analytics that are based on available sensor data, but which provide individualized models that show change over time.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.