ÐÓ°ÉÖ±²¥¼ä

Led by Behnaz Ghoraani, Ph.D.

REU Scholar: Asia Besant

REU Scholar Home Institution: Wayne State University

REU Mentor: Behnaz Ghoraani, Ph.D.

Early Detection of Alzheimer's Disease and Related disorders through Deep Learning-Based 3D Pose Estimation, Gait Analysis, and Machine Learning

Alzheimer’s Disease (AD) is a progressive neurodegenerative disorder with no cure, affecting cognitive and functional abilities in older adults. Mild Cognitive Impairment (MCI) is an early stage of AD. Traditional methods for assessing cognitive impairment—such as lab tests, brain imaging, and neurological exams—are time-consuming, costly, and often subjective. This research presents a novel approach to detecting MCI using deep learning-based 3D pose estimation for gait analysis combined with signal processing, statistical analysis, and machine learning (ML). In this study, we analyzed the curved path walking of 53 older adults, including 27 healthy controls (HC) and 26 with MCI. Our methodology involves several steps: body joint detection using a 3D pose estimation model, signal preprocessing and feature extraction, identifying significant features for distinguishing MCI from HC, and automatic classification using support vector machines (SVM). Our results demonstrate that 3D pose estimation models can accurately detect body joints from video footage. We identified 15 significant gait features, narrowing down to 10 with the highest predictive power for MCI detection. The ML model classified participants with an accuracy of 75.36%, sensitivity of 70.00%, precision of 78.26%, and F-score of 73.47%. In conclusion, our approach—combining gait analysis, statistical methods, and ML—offers a faster, cost-effective, and accessible tool for early detection of cognitive decline due to MCI, suitable for both clinical and non-clinical settings.

REU Scholar: Ethan Zhu

REU Scholar Home Institution: University of Florida

REU Mentor: Behnaz Ghoraani, Ph.D.

Health and Behavior: Next-Gen Health Monitoring Empowered by Python ÐÓ°ÉÖ±²¥¼äming and Deep Learning Applications

Alzheimer’s disease is a terminal, neurodegenerative illness that affects 7 million individuals in the USA, with numbers expected to inflate to almost 14 million by 2060. Due to the incurability of the disease, many studies, like this one, are focused with methods for preliminary detection of the disease, so that early action can be taken to improve quality of life down the line. In this study, we aim to utilize different signal processing and machine learning algorithms on electroencephalography (EEG) signals to aid in the early detection of Alzheimer’s. The dataset utilized for this project included 36 Alzheimer’s, 28 control, and 22 frontotemporal dementia patients, sampled using 19 brain electrodes across a range of times (5.1-21.3 minutes). Extracted signals were preprocessed on EEGLAB, an extension of MATLAB, before being processed using multivariate empirical mode decomposition (MEMD) and power spectral density (PSD). From these processed signals prominent features such as the Higuchi Fractal Dimensions, logarithmic band power, and lempel-ziv complexity were extracted for machine learning. Machine learning models used in this project generated statistically significant results indicating that the methodologies utilized within this project show promising outcomes towards preliminary Alzheimer’s detection. This research has also generated an alternative method of signal processing for future use: noise-assisted multivariate empirical mode decomposition, capable of processing n-dimensional signals for multi-dimensional relationship analysis which can be found online.