Biomedical applications

According to the World Health Organization, stress is any form of change that causes physical, emotional, or psychological pressure [1]. Stress can lead to emotional and mental symptoms such as fear, anxiety, sadness, panic attacks, and depression. Additionally, stress can lead to physical symptoms, including elevated heart rate, difficulty breathing, disruption in sleeping patterns, change in eating habits, difficulty concentrating, and worsening of pre-existing health conditions. Chronic stress increases the risk of developing major health conditions and diseases such as diabetes, depression, heart disease, and cancer. Monitoring an individual’s stress level regularly can help us identify high-stress situations, enabling us to implement early stress interventions for better management. Accurate stress monitoring can be provided through measuring cortisol levels using blood, urine, or saliva fluids. These methods require a specific laboratory or hospital to analyze the sample and determine the cortisol level. These methods, which can be considered minimally invasive, provide a brief glimpse of cortisol levels at a particular moment, restricting their use to continuous long-term stress studies due to their cost and the inconvenience of requiring an external facility to provide a quantitative measurement. Traditional stress detection methods are based on questionnaires in which subjects answer a validated set of stress-related questions to assess their stress level. Nonetheless, these questionnaires cannot be used for instantaneous stress detection. Real-time stress detection methods have been accomplished through analyzing imaging-based approaches that monitor any changes in individual’s facial expressions, including changes in the blinking rate, pupils, and eyebrows. However, these methods could be unreliable due to participants’ manipulated answers and facial expressions regarding mental stress. Despite all efforts reported in the literature, our community has not agreed yet on the number of features to identify stress conditions using wearable devices. The number of used features ranges from 7 to 75 among the reported works in the literature. In addition, most works reported in the literature are focused on individual training and testing. Stress varies from person to person; therefore, a stress detection model based on one subject might not provide accurate stress detection with another subject. An effective stress detection model should provide accurate stress classification for any individual in our society. Thanks to our community’s broad acceptance of wearable devices (i.e., people wear wearable devices during their daily activities), this work investigates a hybrid global stress monitoring framework with generic training and person-specific testing.

Stay tuned if you want to have more information about this research topic. More results are coming soon!

Motion capture systems are widely used for measuring athletic performance and as a diagnostic tool in sports medicine. However, most motion capture systems in the current market have three major drawbacks which are limited portability, affordability, and accessibility. To overcome the drawbacks, this research project focuses on evaluating a markerless motion capture system using a single sports camera and the OpenPose human pose estimation algorithm. The OpenPose algorithm is a pre-build machine learning algorithm developed by the Robotics Institute of Carnegie Mellon University. To validate its accuracy, we have performed a series of tests and evaluations, including running and vertical jumping. We have used the pose estimation from OpenPose to perform gait analysis on our captured running data and compared the results with a marker-based motion capture Qualysis system with 8 cameras. Our results show that the gait analysis results from the single camera and OpenPose are similar to the ones provided by the commercial Qualysis system. Overall, these preliminary findings demonstrate that pairing  low-cost sports cameras with artificial intelligence algorithms allows for high-quality analysis of human movement. 

Stay tuned if you want to have more information about this research topic. More results are coming soon!

Diabetes is a major health concern with approximately 14% of Americans and 9.3% of the global population having been diagnosed with metabolic disease. The most common treatment of diabetes is insulin supplementation. Therefore, accurate and continuous measurement of blood glucose (BG) concentrations is paramount for optimal patient care; however, all current methods of blood glucose measurement and tracking are invasive, requiring skin puncture to evaluate glucose concentrations via blood or interstitial tissues. These current invasive methods limit tracking BG concentrations in some of the most vulnerable populations including adolescents and older adults. We propose to develop and validate a non-invasive wearable (“smart”) device for continuous glucose monitoring (CGM) using an integrated approach of readily available sensors and artificial intelligence. Key features of the proposed wearable CGM device are: (1) completely non-invasive; (2) automatic BG readings; (3) prediction of future BG concentrations to allow for pre-crisis intervention; (4) customizable (user-defined) alerts for changing BG concentrations. No existing system offers users this set of features. Further, the proposed wearable device will have a significant positive impact on the quality of life for individuals with diabetes by offering a non-invasive method of tracking BG concentrations, allowing optimal treatment and health outcomes.  The Clark error plot shows our preliminary results on predicting BG from the proposed non-invasive devices. These predicted BG values have been estimated after training an artificial intelligence (AI) model with our unique set of features (e.g., inputs). The reference/true BG levels were provided by the Freestyle Libre continuous glucose monitoring (CGM) device. From the Clarke error grid (CEG) analysis, 100% of the readings fell in the clinically acceptable zones A.  Although further work is needed to further validate our wearable device, including increasing the number of examples and individuals, it seems that the integrated approach, combining multiple noninvasive wearable sensors to monitor BG presents a promising approach.

Stay tuned if you want to have more information about this research topic. More results are coming soon!

OCT for monitoring the respiratory epithelium: Mucociliary clearance (MCC) is the self-cleaning mechanism of the respiratory tract essential for the prevention of lung infections. When the MCC mechanism is defective, which can be associated with pulmonary diseases Cystic Fibrosis (CF) and Chronic Obstructive Pulmonary Disease (COPD), there are severe and devastating consequences for lung function. Mucus proteins create a matrix-like porous mesh that traps dangerous pathogens to protect the airways. When mucus becomes dehydrated due to disease, it collapses the MCC's cilia lining and pathogens are not cleared from the lungs. In order to monitor disease and determine treatment options, we must know more about the MCC system and the nanostructure of mucus in this dehydrated state.

If you want to have more information about this research topic, visit: https://users.physics.unc.edu/~aold/AppsRespDisease.htm  

Diabetes is currently the world’s fastest-growing chronic disease and it is caused by deficient production of insulin by the endocrine pancreas or by abnormal insulin action in peripheral tissues. This results in persistent hyperglycemia that over time may produce chronic diabetic complications. Determination of glycated hemoglobin level is currently the gold standard method to evaluate and control sustained hyperglycaemiain diabetic people. These measurements are currently made by high-performance liquid chromatography, which is a complex chemical process that requires the extraction of blood from the antecubital vein. To reduce the complexity of that measurement, we propose a fully optical technique that is based on the fact that there are changes in the optical properties of erythrocytes due to the presence of glucose-derived adducts in the hemoglobin molecule. To evaluate these changes, we performed quantitative phase maps of erythrocytes by using telecentric digital holographic microscopy. Our experiments show that telecentric digital holographic microscopy allows detecting, almost in real time and from a single drop of blood, significant differences between erythrocytes of diabetic patients and healthy patients. Besides, our phase measurements are well correlated with the values of glycated haemoglobin and the blood glucose values.