Smart health systems (SHSs) offer continuous vitals monitoring, providing easier access to effective treatment, reducing cost and delay, and, most importantly, decreasing patient mortality by ensuring non-disruptive medical services and early detection. Nevertheless, deploying SHSs is impeded due to security and reliability issues from the susceptibility to state-of-the-art exploits and zero-day attacks, especially on IoT/sensing systems. In addition, portable sensors are susceptible to significant error due to unrestricted user activity and limited human testing, as exposed in common pulse oximeters during the pandemic. The limitations of portable sensing can be overcome with long-term monitoring and multi-modal sensing, which requires onboard AI to efficiently and securely collect high-quality data. Long-term data can then be used to develop predictive and personalized models for actionable real-time measurements. This work looks to improve the development of existing SHS by exploring how to implement novel health solutions leveraging advanced sensors and AI. This REU explores advances in fields such as motion health analysis, cardiac disease detection and mitigation, mental fatigue and stress,  environmental changes and analysis, and spectroscopy detection.

Projects Overview

Adaptable Optical Monitoring System Design

One of the main challenges with wearable devices is the limited palette of detectable biomarkers and measurement types due to noise. To mitigate this challenge, this project explores UV-Vis measurement alternatives in portable devices, significantly limiting the noise present in the system. In this project, students will collect measurements of different light sources on different organic and inorganic films to design the rules for the other light measurement detectors.

Multimodal Wireless Sensor Fusion with Embedded Packaging

Existing health-monitoring modality sensors work independently from one another, which limits their ability to identify systemic level issues. This research project looks to integrate multimodality sensing such as Electrocardiogram (ECG), or Seismocardiography (SCG) alongside machine learning for cardiac disease analysis. Students in this project will develop autonomous wireless multimodal sensors with embedded component packages for communication.

Multimodal Sensors for Continuously Movement Measurements

Recent advancements in technology have led to increases in health tracking. One key area that has seen a rise is tracking movement and weight to combat the current obesity epidemic. However, multiple sensors have to work in tandem for effective movement measurement. In this project, students will design multimodal sensors for movement measurement without requiring intrusive amounts of devices.

IoT Infrastructure Development for Smart Health Systems

 Traditional Smart Health Systems for analyzing brain activity utilize Near Infrared Spectroscopy (NIRS). NIRS sensors typically require using scalp caps, which are bulky and expensive and lack additional information on the patient being examined. This project looks to design novel SHS for NIRS that are nonintrusive and multimodal. Students in this project will design novel devices and implement data analysis techniques to develop models.

Gopher Tortoise and Biodiversity

Gopher tortoises are a federally protected species local to the South Florida ecosystem. A data-driven approach is required to improve the effectiveness of conservational efforts for these environmental systems. In this project, the aim is to develop a comprehensive system to monitor tortoise movement and vitals. Students will assist in developing the sensors and implementing ML models for onboard data processing.

Multiplexed Volatile Organic Compound Sensor for Breath Analysis 

In nature, receptors typically handle odor analysis, where each receptor identifies multiple types of molecules. In this project, we look to design susceptible and volatile organic compounds to detect diseases at an early stage. We will use the sensor readings to develop advanced AI/ML models to detect early disease. Students will work with different volatile organic compounds and identify and develop artificial neural network models.

Toward Resilient Smart Healthcare Control Systems

SHS leverages body sensors to collect patient health information. It processes health information using AI/ML techniques to operate medical devices. However, communication side SHS are vulnerable to adversarial attacks. As a result, it is integral to implement security and resilience in SHS to guarantee continuous and effective health monitoring and control. In this project, students will design multi-fusion SHS controllers, anomaly detection modules, and Generative Adversarial Network models for more comprehensive training.

Formal Attack Analytics for ML-Driven Smart Healthcare Systems

In safety-critical systems such as SHS, manipulated sensor readings can result in increased threatening events. As a result, it is crucial to analyze the vulnerabilities and robustness of SHSs. One such explored way is to leverage adversarial ML techniques to find attack vectors. However, this is not verifiable. A formal attack synthesis tool produces verifiable attacks which are then defensible. This project will look to find the formally verifiable attack vectors on SHS. In this project, the students will identify adversarial ML attack vectors, design formal threat synthesis tools, and identify formal threats for SHS.