Hero: Novel Approach to Combat Substance Abuse Disorders via AI-based Emotion Inference, GNP-based Metabolite Detection, and Novel Biometric Measurement
Over the past year, approximately 10,000 American teens have died due to a prescription stimulant overdose, and over 1.7 million teens suffer from a substance abuse disorder. We currently lack a way to accurately and noninvasively detect overdose behaviors, prompting the question, can a multifactor system be built to detect a prescription stimulant overdose? The eight factors researched in this study include mood swing detection, sweat composition analysis, measurement of four vital signs, detection of spasms, and consciousness verification. The overarching research question was split into three queries: can a mathematical model quantify the relationship between melancholic and euphoric diction to detect a mood swing (hypothesized a significant difference in mean score), can a reagent strip be designed to exhibit a difference in color when in contact with healthy sweat versus sweat containing simulated amphetamine metabolite (hypothesized a significant color change measured by a spectrophotometer), and can a PPG algorithm be built to noninvasively estimate vital signs (hypothesized 85% accuracy). Three types of Artificial Intelligence models trained on over 1.6 million data points were successfully able to detect a mood swing (maximum AUC of 0.86). The final model was implemented into the Hero mobile app to monitor a teen’s outgoing SMS messages for a mood swing. The gold nanoparticle-based reagent strip displayed a significant color change from red to purple when in contact with amphetamine metabolite; an apparatus was built to integrate the strip into a teen’s daily life. The vital sign algorithm resulted in a percent error of <5%.