Hero: Automated Detection System for Prescription Stimulant Overdose via AI-Based Emotion Inference, Metabolite Detection, and Biometric Measurement

作     者：Divya Nori Divya Nori

作者机构：Milton High School Milton GA USA 

出 版 物：《Open Journal of Applied Sciences》 (应用科学（英文）)

年 卷 期：2020年第10卷第12期

页      面：791-816页

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：Addiction Science Behavioral Science Artificial Intelligence Biochemistry 

摘      要：Over the past year, approximately 10,000 Americans have died by psychostimulant overdose, and over 50% of these deaths were caused by prescription stimulant misuse. A comprehensive approach to detect a drug overdose in the environment where it occurs is imperative to reduce the number of prescription stimulant overdose-related deaths. Teenagers are at the highest risk for prescription stimulant overdose, so this study proposes a multi-factor overdose detection system named Hero which is designed to noninvasively operate within the context of a teen’s life. Hero monitors five factors that indicate stimulant abuse: extreme mood swings, presence of amphetamine metabolite in sweat excreted from the fingertip, heart rate, blood pressure, and respiration rate. An algorithm to detect extreme mood swings in a teen’s outgoing SMS messages was developed by collecting over 3.6 million tweets, creating groups of tweets for euphoria and melancholy using guidelines adapted from DSM-5 criteria, and training six Artificial Intelligence models. These models were used to create a dual-model-based extreme mood swing detection algorithm that was accurate 96% of the time. A biochemical strip, which consisted of a diagnostic measure that changes color when in contact with amphetamine metabolite and a control measure that changes color when the appropriate volume of sweat is excreted, was created. A gold nanoparticle-based diagnostic measure and pH-based control measure were evaluated individually and on the overall strip. The diagnostic measure had an accuracy of 90.62% while the control measure had 84.38% accuracy. Lastly, a vital sign measurement algorithm was built by applying photoplethysmography image processing techniques. A regression model with height, age, and gender features was created to convert heart rate to blood pressure, and the final algorithm had an accuracy of 97.86%. All five of these factors work together to create an accurate and easily integrable system to detect o