Parallel Patients

When I shadowed an Orthopedic Pediatric surgeon in Minnesota before coming to med school, I got a chance to observe a very prestigious and experienced doctor interview a young girl with a very rare congenital condition in her leg. He came out of the clinic with a resident and myself trailing him and he showed us a binder he had with case summaries in it from his past and he opened it right to a case which was very similar when a different kid had the same thing and it talked about the case and what was done. I was impressed, and in Medical School as a side project I have been working on a project to give any doctor the same bounty of experiential knowledge that this prestigious doctor had, and even more. I envisioned a tool in an EMR where once you put in your observational note of a patient who had come in, you could click “Find patients like this” and see a list of the top 10 most similar patients the hospital has ever seen, what their doctors did, and how it turned out. It wouldn’t tell you what to do, but you would be able to have some perspective and distance yourself from the biases inherent in your teaching, specialty or history. I learned recently that the biggest predictor of what drugs a doctor prescribes is what they had used previously. This makes sense, but is it the best thing for patients? There are thousands of drugs on the market, it seems like some collective experience and shared advising could help patient care.

I couldn’t have begun work on this project without the availability of the MIMIC-III research database of real ICU records from 40K+ patients from the Beth Israel Deaconess Medical Center, anonymized by hand to remove personally identifiable information. I discovered this dataset when I competed in the HST Critical Data hackathon focused on novel uses of it, and my friend Sean Wang and I won with the prototype of this we called Parallel Patients! I’ve continued working on this project ever since, maybe too much, and I have been learning a lot of interesting medicine through analyzing specific cases and asking myself what really makes two patients meaningfully similar. The good news is that it has been measurably increasing in capability over time and has been showing some very exciting results! The latest work I’ve done with it is to predict a patient’s future by looking at the future records from parallel patients of the given patient. What’s really cool is that you can test these predictions by taking an early record from someone in the database, naively predicting the future of the patient, and then comparing that with their own future records. I’ve even been optimizing the weights of the algorithm to learn what similarity factors are most important in making this prediction. Very exciting work!

Also, recently Parallel Patients qualified as a finalist in the Massachusetts Medical Society IT Award, and on Dec 7th I will present it to a crowd of doctors! I’m excited to hear their feedback and to polish it up so that it can be as good as possible by then.

Dr. Memory App

Many things that I have been learning in medical school have been opening my eyes a bit more to what really goes on in diagnosing a patient. One of the surprising, but in retrospect obvious, things for me has been how important the interview and past medical history is to making a diagnosis, even compared to lab results and medical images which I had thought were more infallible than they really are. We learned about studies like this one which concluded that medical histories are 60% of the important source of data for diagnosis with the physical and lab results being the other two 20% respectively. What shocked me about this is how in my own observations and practice interviewing patients, I know that I’m only able to jot down a fraction of the interview and taking notes is pretty much inversely correlated with the strength of the patient-doctor relationship which is very important. I know that people over-estimate their own recall ability, which normally isn’t as big of a deal, but when the interview is clearly super important why are we forcing all the info to go through a bottleneck like the human memory, when we could record it with audio for a doctor or scribe to use or go back to later to refresh their memory. Little things mentioned off hand have been often shown to be important signs that could really help guide a differential diagnosis when seeing a blob on an x-ray which could mean many different things.

In order to brainstorm how this could be put into practice as a memory aide while maximizing patient experience, ease of use, and security, I first came up with a clip on microphone attached to a phone and a wireless remote in my pocket for bookmarking when important things are said for going back to later. This ended up being a bit too many pieces to manage and bookmarking wasn’t as intuitive as I had thought, oh well good thing it was quick to test with a friend! I changed my plan and decided that a good flow would be to come into a room and ask “Is it okay if I record this conversation for my own memory?” (and fall back to paper if they say so, easy) before putting a dedicated cheap phone down on a table which I can tap to start recording with my own app. I just finished this version of the app and successfully got it to encrypt audio and retrospective bookmarks and names with a password that it asks for every time you want to review a recording. The phone also isn’t connected to the internet at all, and all files are encrypted. I’m excited to begin using it as a memory aide with real patients once I get approval!

ESP: Private Wearable Notifications

I’ve had a lot of cool opportunities to try living with wearable devices like Google Glass, Android Wear, and even making my own makeshift wrist-worn phone many years ago. I’ve seen the benefits and inconveniences in each of them, and the feature that I have found to be the most useful of them has been timely notifications. Being hands-free aware of your environment and the world can empower your decisions and guide your live in positive ways. However, some devices are socially intrusive (Glass) or give off the wrong message to others (checking a smartwatch when it buzzes makes people think you are late or bored). I decided that my ideal smart notifier would almost be like a brain implant in which you instantly knew when breaking news happened, or your friend asked for you to meet, or if was about to rain, even if you were riding a bike.

I didn’t make a brain implant, but I did find a small bluetooth earbud which my phone can talk to me through and does everything I wanted! It’s so easy and awesome I’m surprised it’s not more popular of an idea these days. I call the software ESP after the classic psychic fake phenomenon of extrasensory perception. I can use any bluetooth headset or a wired headset and I keep adding new features to empower my daily experience. Recently I was biking and I heard a voice in my ear that said “Facebook message: Rebecca. Hey I just saw you go by!” and I stopped and went back to see my friend. Also, while working I can listen to music and every hour the music gets quieter and says “Your first task is shop for a bike and the time is three o’clock” because it is tied to my task list. Biking turn-by-turn directions have also made my commute super enjoyable. People also have surprisingly not noticed it much, especially my new one which goes behind my ear. I do get these moments where I feel like I have super power or something when I’m with some friends and someone messages our group message on facebook or hangouts with an update and I share it without missing a beat or pulling out a phone “Oh, sound like Nick’s on his way.” This is fun. Also, I have CNN, The Onion, The Verge, and Elon Musk’s twitter stream read to me, which is fun. It may sound like a lot, but these are really short messages and easy to ignore. I set it up too so that if you ignore them during a conversation (easy to do in one ear) then you can tap the side to toggle through the history. I can also tap at any time to hear the time and the song playing if there is one. No need for a watch any more!

My app has been pretty customized to my own needs (my task list and preferences for weather, twitter, and news feeds) so I haven’t worked on publishing it. There are cool apps out there already for speaking See Voice Notify and let you make sure that it doesn’t speak when headphones or bluetooth headsets aren’t connected (awkward).

In a broader sense I could see this being a great tool for teams collaborating, and even medical professionals who could get pager emergency messages discretely spoken to them to allow them to not have to interrupt people around them with their own beeper loudly going off. I’ve also seen people working in shopping store and security guards with similar discrete headsets, and I think they aren’t the only people who could benefit from them. Now if only you could send a message by thinking.. I did already make a wireless bluetooth remote that I can send texts to people and queue up music without looking with, but it relies on my knowledge of morse code input which I don’t expect other people to learn haha.

vOS: a Virtual Operating System for Virtual Reality

At the Google I/O Research Lab Conference, I presented an open source project I have been working on called vOS which stands for "virtual Operating System". vOS started with a question: "What is the richest interface for programs designed for an immersive heads up display?" Game pads and gesture interfaces are bad at text input and fine control, keyboards are difficult to use when you can’t see your hands. This became a continuously evolving experiment to make a very intuitive remote control tablet interface for people wearing virtual reality headsets.

See the poster Try it Out See the Code

Video Embed Chrome Search Extension

This was a quick chrome extension I wrote up when I noticed that sometimes your Google search results give you a nice big video result that looks like it has a play button, but actually just links you to YouTube. This extension detects that and embeds the real video, saving you a small but nice step and lets you continue to look at your search results.

Try it out

LEAPnav website element

LEAPnav is a website element which lets people easily control a website using hand gestures and a LEAP motion device or using the keyboard as a fallback. LEAPnav is used on this website on the bottom left and hopefully other websites will use the open sourced code. The most powerful part of the plug in is the intuitive visual feedback which helps the user guide their hand. As gestures are formed, the display shows what is being registered in order to teach the user how to give clear signals. Four screenshots of this in action are shown on the left.

Get the Code on Github

Link Toss

Link Toss is a project I did to learn more about web connected apps and to solve a need that I had encountered. Essentially, it is a frictionless way to share links (URLs) to people who you are currently with. You open the app or website (Android for now, iOS soon) and paste your link or click share from the browser and once you click "Toss" then it gets tied to your geo-location and is made public on the website for 4 hours. Also, anyone with the app installed and who is within a radius that they set will immediately get a notification with the link on their phones.

In practice, when I'm with a group of friends going out, I can toss a link to where we are going or to an article I'm talking about and everyone gets a notification with the info. I could see this being useful in a classroom or even fun at a concert! You can try it out now at bit.ly/linktoss!

Download Android App Try The Web App

Custom Home/Now Launcher

I originally made this app as a way to made the google search engine replace my homescreen on my android phone (show up when I press the home button) but it has evolved to be much more. The original app was called Google Search Launcher and received 186,589 installs and 4.25 stars out of 767 reviews. I have since replaced it with two apps, one called Custom Home Launcher and another called Custom Now Launcher. Each lets you choose any one of your apps to launch when you press the home button or drag the home button up respectively. I put a lot of work into making sure that people wouldn't have trouble changing their home screen or Now launchers back to the way they were before, so whenever the app is used a notification comes up which allows the user to change their settings.

Check Out My Android Apps

Q & A Hangout App for Teachers or Gameshows

To prepare for my Google Internship on the Hangouts Platform team, I wrote this Hangout App using the Hangouts API to allow people to take polls in a Hangout and raise their hand with a hangout overlay. People can also earn "points" for getting answers correct and a person's points show up by their window in the hangout. The raised hand feature, points overlay, and who is the manager of the app can all be modified.

I think this app would be great for a Hangout classroom experience to test audience members on how well they are comprehending a lecture as well as a way for student to raise hands if they have questions. Alternatively, this could be used in a Hangout gameshow to award points to people who buzz in with their hands first and get the right answer.

Check It Out!

Dining with Devils at Duke

At Duke, after freshman year is over, it has been hard to expand your social circle past your existing group of friends or people in your grade or major. I wanted to help combat this by fostering an environment which is open for meeting new friends from any class or major. Many of my friends also were interested in getting involved and we started Dining with Devils, a group which plans dinner times and places for people to meet strangers at Duke.

View the Website

Duke Notes

In my freshman year of Duke University, I started something called Duke Notes which became a huge hit! It all started in my intro to engineering class when I made a collaborative Google Doc for taking notes and got a hundred people in the class to take notes together in real time. I wanted to make it easier for people in the same classes to get in a Google Doc with their classmates so I make the Duke Notes website where people could find documents for their classes. It exploded to be used by 300 people and very high quality notes were being created quickly with beautiful images from around the web and helpful comments by peers. Teacher assistants were using it to answer questions or fix mis-understandings in real-time and teachers were talked to about their opinions about the service. The site has since stopped, as it seems to have worked best for large scale introductory classes which don't hand out pdfs of lectures beforehand. I am glad that at the time it served a real need and helped out a lot of people.

See Duke Notes from 2012

Adventive Mobile Web Browser

This is a web browser for android that is built for web apps. All webpage actions are in a bar that comes up when you click the menu button. You can have multiple tabs and you can duplicate tabs while maintaining the history of each tab. History can be accessed with a menu button press in the pages activity. History is kept in a running list which can be fully cleared or individual items can be removed.
Additional Features:
- Any webpage can be added to your home screen as a shortcut by "pinning" in in the browser pages list
- You can put an icon on your screen for both the main app as well as a shortcut to the pages open.

Get the App

Fully Submersible Water Turbine

In my senior year of high school in 2011, I had an idea for a water turbine design which could work completely submersed in water flow as opposed to traditional water turbines which only have part of their turbine submersed in the water flow. I developed the design and ran mathematical simulations and built a lego prototype before building a scale model out of household materials. I was able to test the power output of this design against the power output of an equally large turbine of a traditional design in a water flume powered by the mississippi river and I found that my model produced significantly more power. While this is important, I believe that the true benefits of my design lie in the fact that the blades act like large sails and the tips move only up to twice the speed of the water as opposed to regular blades which increase in speed with blade length. This reduces the chance of cavitation which wears down on traditional turbines. Also, this turbine works in both directions naturally and could be used to collect energy from tides.

I presented my project in the 2011 Intel International Science Fair in Los Angeles, California along with 1600 other fellow science oriented students from around the world and ended up winning a 4th Place Grand Prize in the ctegory of Energy and Transportation.

I also entered my project in the first ever Online Google Global Science Fair. In this fair, unlike how I've normally presented in person with a posterboard, I had to make an online poster/website for international judges to check out. I ended up being selected as a one of five finalists in my age category and I was flown out to the Google Headquarters in Mountainview, CA for the GSF Finalist Event! I had colored my model turbine Google colors by the competition and I think they liked it.

Explore my Online Google Science Fair Poster

The Circuit Head Accessibility Device (CHAD)

In my junior year of high school in 2010, I was researching Brain Computer Interfaces and was amazed by how a small brain implant could one day allow people with no motor function to be able to control the devices around them. I was put off by how far off this dream seemed so I began brainstorming how some of these benefits could be achieved today through alternative means. I then came up with the idea for the Circuit Head Accessibility Device (CHAD) which is a hat that can sense tilts in a user's head using an accelerometer and sends commands to a computer via a USB cable to control a computer cursor similar to a joystick. The user can click by biting on a bite sensor attached to the hat.

I developed this device from parts ordered online and built it in my bedroom and programmed it to be able to calibrated to anyone's natural resting head position and not be too sensitive to small natural head movements. I also visited the Courage Center rehabilitation facility where I got to get feedback from a dozen people who used assistive devices on a regular basis, and I was excited to find out that this device was much more easy to use than alternative devices!

I designed a mouse task program with shifting targets which could test how fast and accurate a person was with a cursor, and had able bodied people try completing the task multiple times with a mouse and then with the CHAD. I found that people got better with the CHAD over time and quickly got up to 30% of the efficiency of a normal mouse! This is a big deal because of all the alternative devices I tested, the max efficiency I found was 10%, which can be very frustrating if that is the only option.

I presented my project in the 2010 Intel International Science and Engineering Fair in San Jose, California and was delighted to receive a 4th Place Grand Prize in the category of Electrial Engineering. I also ended up winning a Davidson Fellowship and was inducted into the National Gallery of Young Inventors because of the CHAD, and met so many amazing people along the way. My current plans are to build the CHAD 2 after gaining more electrical engineering experience and the new version will be wireless.

Articles about the CHAD have been posted in CNN, the Star Tribune, and Minnesota Public Radio!

WatchOut 3D

Augmented Reality (AR) and computer vision have interested me for a while, and I have been brainstorming how AR can become more than just used to view static 3D models, but become an interactive experience. One of my projects to explore this was Watch Out, an app which is meant to be used with a paper watch (with a latch mechanism which I developed). Now, the paper watch can't tell time, but when it is observed through the phone app, the watch hands appear in their proper place and the digital time appears to be floating and circling the watch. The best way to see this is by watching the video demo I made.

Download The Android App Print Out Your Own Watch

MathDroid For Kids

MathDroid for Kids was one of the first mobile apps that I made, and I made it for my sisters who were learning elementary math at the time. This was right before Khan Academy became big, and I thought it would be really cool to have an app which could algorithmically not only come up with practice math problems of a wide range of types, but also provide challenging alternative multiple choice answers and also adapt to your ability over time by providing problems at your level of ability. I also added in a point system and a mini store where kids could use points to buy accessories for an android character as well as earn badges by beating boss charachters for subjects like addition, subtraction, multiplication, and fractions. I was relaly happy to hear feedback from parents who used it to help their kids learn math! I made a section for showing overall progress just for parents. The app was built for smaller phones at the time, and hasn't scaled well to larger devices, so I know I need to go back and fix that.

Download The Android App

Room Guide

In high school, I was just learning Java for the first time when I wanted to apply my knowledge to make a big difference. While I had no initial knowledge of path finding algorithms when I started, I set to work building a java application to help guide students and visitors around my giant three floor high school built to handle 3000 students. I spent a lot of time mapping out the school from the maps given to us in our planners and I now know the fastest way to get any where in the building, but in the end it all came together and the app was finished. IT was installed on all of the computers in my school and achieved its goal of helping students and visitors get around the school! Some cool features include, draggable start and end points, optional elevator usage, pop-out step by step navigation with landmarks referenced and printable results with augmented maps across levels along with the written directions. I discovered Djikstra's Algorithm in the process of writing this code, and I gained a lot of valuable experience designing user interfaces.