INNOVATIVE FENTANYL DETECTION: THE INTEGRATION OF DEEP LEARNING WITH MOLECULARLY IMPRINTED POLYMER SENSORS
Restricted (Penn State Only)
- Author:
- Darweesh, Raheeq
- Graduate Program:
- Electrical Engineering (MS)
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 22, 2024
- Committee Members:
- Nashwa Nabil Elaraby, Thesis Advisor/Co-Advisor
Seth Wolpert, Committee Member
Rafic Bachnak, Professor in Charge/Director of Graduate Studies
Aldo W Morales, Committee Member - Keywords:
- Fentanyl
Breathalyzer
MIP
LSTM
RNNs - Abstract:
- The growing misuse of Fentanyl, a highly addictive painkiller prescribed for severe pain, has emerged as a significant public health crisis. With over 110,000 drug overdose deaths in 2022, Fentanyl was responsible for 70% of these drug overdose deaths, making it the leading cause of death for young Americans aged between 18-45 years old [1][2]. The brain's response to Fentanyl is similar to heroin and thus was unfortunately misused beyond the intended medical purposes [1]. Considering this critical issue, this thesis addresses the importance of accurate and fast Fentanyl substance detection. The research uses a novel molecularly imprinted polymer (MIP) sensor that can detect Fentanyl in a human breath or an enclosure, configured as a breathalyzer. The MIP sensor consists of three layers, a silicon substrate layer, a graphene-Prussian blue layer, and a MIP layer. The selectivity and specificity of the MIP for Fentanyl molecules allow for high-fidelity detection, aiming for a backdrop of the escalating Fentanyl misuse crisis. This research prioritizes the development of a highly accurate and reliable algorithm by employing deep learning techniques, specifically Long-Short Term Memory (LSTM) model, a type of Recurrent Neural Network (RNN). LSTM is renowned for its accuracy in pattern recognition and prediction in time-series data, making this technique ideal for analyzing the dynamic responses of the sensor to trace Fentanyl presence. This research emphasizes not only the accuracy and reliability of the detection algorithm but also its operational efficiency. Results are delivered within a 30-second window, divided into three critical phases: 10 seconds for sensor calibration to ensure baseline accuracy, 10 seconds for the breath exposure test capturing the sample, and the final 10 seconds dedicated for sensor's post-exposure behavior. The model was trained based on 162 samples: 90 were real positive tests (Fentanyl was present), 35 were real negative tests (Fentanyl was not present), and 37 were synthetic negative tests (Fentanyl was not present). The trained model has shown an accuracy of 99%, highlighting the system's potential for real-world application.