Mobile Health Systems

Smartphone-Based Ischemic Stroke Screening

Every year over forty thousand strokes are caused by internal carotid stenosis in the US alone. Our research aim is to identify patients that are best suited for carotid artery surgery for stroke risk detection through facial videos on a mobile phone. I am applying computer vision and signal processing techniques to extract the pulse of a user from facial videos. A delay in the pulse between the left and right sides of the forehead indicates possible blockage of the carotid artery.

Mobile Ultrasonic Sonar Exercise Sensing

Physical inactivity is the fourth leading risk factor for death worldwide, and yet eighty percent of US adults do not meet national exercise recommendations. Our aim is to quantify upper body movement using a novel sonar sensing method on a mobile phone. Though FitBits and similar technologies count steps using accelerometers, such devices are prohibitively inadequate for many obese, elderly, or disabled populations. With a fellow intern, I helped build a health application to track upper-body exercises, targeting millions with sedentary lifestyles. I coded a data collection web application, helped conduct a study, and developed repetition counting techniques. We additionally applied deep learning in Python to classify over 10 types of physical activity.

Astrophysics

Stellar Photometry

Asteroid Orbit Determination

In the summer of 2020, I was thrilled to be 1 of 36 students selected internationally to participate in the Summer Science Program in Astrophysics. Over the course of five weeks, I spent over 300 hours working with a three-person observing team to remotely operate a research-grade telescope to obtain images of the near-Earth asteroid 2011 XZ1. We performed photo reduction and astrometry on the images and used the asteroid's coordinates to calculate its orbital path, including the chance it will impact Earth in the future.

Computational Musicology

Classifying Raags

On my Junior Diploma exam in Hindustani Classical Music, I struggled with the challenge of differentiating between raags, the name given to Hindustani equivalents of Western melodic scale patterns. Each raag is distinguished by emphasis on certain notes or re-occurring patterns. One day, as I was practicing for my exam, I thought to myself - could a computer possibly complete this classification between raags that I struggled with? The architecture of my lab research’s machine learning network immediately popped into my mind. I embarked on a year-long individual research project where I coded a one-dimensional convolutional neural network to classify different raags. My IB Extended Essay on this research topic can be found below.