Student Horace Greeley High School Chappaqua, New York, United States
Introduction: Millions of people in the United States suffer from Alzheimer’s Disease (AD), an incurable form of dementia that continues to increase in prevalence. Current methods of AD diagnosis are limited to the late stage by which time the treatment options are limited, quality of life is poor, and cost of treatment is exponentially high. However, early medical diagnosis of AD is difficult since standard non-invasive techniques require extensive tests and can still generate false positives and negatives, leading to misdiagnosis.
Materials and
Methods: This study proposes a supervised machine learning model trained on readily available Electroencephalography (EEG) patient data to diagnose potential AD patients. This study was conducted by extracting and analyzing relevant features from an open source EEG database, collected from 186 patients using the trained machine learning model of best fit. This AI model is an alternative to current late-state detection methods which require complex and risky procedures that can still lead to inaccuracies. In addition, current algorithm require feature manipulation and sort through hundreds of thousands of raw EEG data points to obtain unreliable results.
Results, Conclusions, and Discussions: The results demonstrate that, given EEG data of 93 close-eyed patients, the trained Logistic Regression model - the machine learning model of best fit - achieved a sensitivity of 100% and overall accuracy of 87%, using data recordings of only 8 second segments for each patient. This novel Alzheimer’s Disease screening tool, with a cloud-based AI model can be easily deployed at primary health care clinics for screening patients for AD during their yearly clinical visits.
