A Mobile Platform for Non-invasive Diabetes Screening (Master’s thesis)

Mobile Technology Group, D-Lab

For my Master’s thesis, I worked on developing a mobile application for type 2 diabetes screening in India, known as the diabetes capital of the world. Early detection is needed to manage the disease, but the demand for health workers and professionals in India far outweighs the supply. The traditional method of screening for diabetes is to test a small amount of blood with a glucometer to check for blood glucose levels. However, blood tests carry a risk of blood-borne diseases and most health clinics in low-resource areas do not have a ready supply of paper test strips to use with the glucometers.                                                 

Our solution, a mobile platform, uses a highly portable mobile app and imaging tools to gather non-invasive sources of information, conduct analysis and deliver near immediate results for hard-to-access populations in urban and rural areas of India. The non-invasive tests include clinical questionnaires, thermal imaging, iris imaging, retina imaging and photoplethysmography. For the purposes of evaluation, we defined six stages of diabetes progression: stage 0 (no diabetes), stage 1-3 (prediabetes), stage 4 (diabetes), and stage 5 (advanced diabetes).

As a sample real-world implementation of this platform, our team began a large-scale diabetes study in two different sites in India, one in Mumbai and one in the Bangalore area. Based on data from this field study, I developed Bayesian network models to predict whether a subject is in one of three different stages of diabetes pathogenesis: stage 0-3 (no diabetes to prediabetes), stage 4 (diabetes) and stage 5 (advanced diabetes). Though data collection for the project is still ongoing, early results are encouraging. My full thesis and results are available here.

Popular YA

One fun project I recently worked on is PopularYA, a website where I review recent books that I've read. This was a really great way to learn more about creating a personal website and the challenges/considerations that go into building a website from scratch. I started with a Bootstrap template that I liked, Avilon, and then customized it to fit my needs. I connected a Mongo database and used Node.js and Express on the backend. The coolest part of this website is that I just have to add my review to the Mongo database and it populates the review onto my website, which is super convenient. I also went through the process of buying a domain and deploying the application through Heroku. I want to add more features in the future such as more reviews, a Hall of Fame section for my favorite books, better analytics, search functionality and so on. I know I still have a lot of work to do on this but please feel free to check out the website at www.popularya.com!

Question Answering

This is a partner project I did for a final project in a class, Natural Language Processing. where we implemented a variety of models, including an LSTM and CNN, to try to find similarities between questions, specifically StackExchanged Android questions. This has real-world importance because one way of answering someone's question is to find whether someone else has asked it before (and use the corresponding answer). The full report we turned in for the final project can be found here.  And the code for the final project is here.

Image Colorization

This project was part of my final project in my Computer Vision class. My partner and I used a Convolutional Neural Network (CNN) to colorize black and white images. The code can be found here. Enjoy some of our results below, the image on the left is the colorized image produced by the CNN and the image on the right is the actual image. The final project paper is here.