AI-Driven Predictive Modeling of Alzheimer’s Disease Progression Using Deep Learning and Clinical Data
Disciplines
Biomedical Informatics
Abstract (300 words maximum)
Alzheimer's disease is one of the most important public health problems of our time, affecting millions of individuals worldwide. As a chronic neurodegenerative disorder, Alzheimer's leads to cognitive decline, memory loss, and, ultimately, loss of autonomy. Our research aims to employ artificial intelligence, through a Dense Neural Network (DNN) model, to analyze the progression of Alzheimer's based on an individual's exercise, diet, lifestyle, and current condition. Our data was obtained from WashU Medicine’s Open Access Series of Imaging Studies (OASIS) database of cross-sectional MRI scans from patients that ranged in age from young to older adulthood. The two datasets that were used focused on showing the progression of Alzheimer’s (cognitively normal, uncertain dementia, and AD dementia) over time in a variety of patients. They were first merged via their SessionIDs to ensure the model’s ability to track a patient’s cognitive progression in case their OASISID had multiple SessionIDs attached to them. The data was then preprocessed to ensure data integrity through scaling, encoding, and handling NaN (null) data values. A Dense Neural Network model with two hidden layers was implemented using TensorFlow and Keras, optimizing for both accuracy and generalizability. The model was trained with a categorical cross-entropy loss function, adaptive learning rate optimization via the Adam Optimizer, and class weight balancing to mitigate bias against important, yet underrepresented classes, for the sake of generalization. The model achieved a classification accuracy of 96% ± 2% after 50 epochs, demonstrating its potential for accurate predictive analytics in biomedical applications. By identifying patterns and correlating disease progression, we aim to generate predictive insight that can be used to support early intervention and customized treatment methods.
Academic department under which the project should be listed
CCSE - Information Technology
Primary Investigator (PI) Name
Chloe Yixin Xie
AI-Driven Predictive Modeling of Alzheimer’s Disease Progression Using Deep Learning and Clinical Data
Alzheimer's disease is one of the most important public health problems of our time, affecting millions of individuals worldwide. As a chronic neurodegenerative disorder, Alzheimer's leads to cognitive decline, memory loss, and, ultimately, loss of autonomy. Our research aims to employ artificial intelligence, through a Dense Neural Network (DNN) model, to analyze the progression of Alzheimer's based on an individual's exercise, diet, lifestyle, and current condition. Our data was obtained from WashU Medicine’s Open Access Series of Imaging Studies (OASIS) database of cross-sectional MRI scans from patients that ranged in age from young to older adulthood. The two datasets that were used focused on showing the progression of Alzheimer’s (cognitively normal, uncertain dementia, and AD dementia) over time in a variety of patients. They were first merged via their SessionIDs to ensure the model’s ability to track a patient’s cognitive progression in case their OASISID had multiple SessionIDs attached to them. The data was then preprocessed to ensure data integrity through scaling, encoding, and handling NaN (null) data values. A Dense Neural Network model with two hidden layers was implemented using TensorFlow and Keras, optimizing for both accuracy and generalizability. The model was trained with a categorical cross-entropy loss function, adaptive learning rate optimization via the Adam Optimizer, and class weight balancing to mitigate bias against important, yet underrepresented classes, for the sake of generalization. The model achieved a classification accuracy of 96% ± 2% after 50 epochs, demonstrating its potential for accurate predictive analytics in biomedical applications. By identifying patterns and correlating disease progression, we aim to generate predictive insight that can be used to support early intervention and customized treatment methods.