Features for LCD
Loading page when LCD is running the diagnosis
Results page for health
Results page for cancer
We identified the lung cancer is one of the most critical diseases, for the chance that a man will develop lung cancer in his lifetime is about 1 in 14 and for a woman, 1 in 17. It is crucial to detect cancer at early stage to ensure timely treatment and increase survival rates.
However, the diagnosis process is time-consuming and prohibitive, which hinders people for actively monitoring their health conditions. Furthermore, the manpower cost to hire a specialist for hospitals is costly in terms of both time and finance, with an extensive training required.
After exploring the elegance of machine learning, we believe that the deep learning methodology can contribute largely towards this pressuring issue. Within 2 days, we built LCD, an innovate deep learning tool dedicated to lung cancer diagnosis.
What it does
LCD accelerates the detection process for lung cancer. Users can simply upload CT scans to LCD website and get the diagnosis results within minutes. It not only detects potential cancer nodules, but also eliminates false positive nodules to ensure higher accuracy.
How we built it
We trained the image segmentation model using neural networks, specifically, U-Net. Then we use the Residual Networks to eliminate the false positive samples. We trained the neural networks with Tensorflow framework. The model is running interactively on a GPU server. A web app is built with Python Flask framework that allows users to have minimum trouble of getting diagnosis immediately.
Challenges we ran into
Time constraint for deep learning is beyond our expectation. Usually, it takes 2 weeks to train a fully functioning LCD, but we do not have the luxury of time to make a complex and sophisticated LCD from scratch within 2 days. It's depressing to realize that the sophisticated version that coders have spent the entire first-day writing might not be able to implement completely. However, we made some necessary adjustments and manage to present the first version within 48 hours.
Accomplishments that we're proud of
We are proud to build a product with strong future applications and positive implications towards the society. Deep learning for cancer detection is such a fascinating field that illustrates the combination of two seemingly irrelevant areas. NYU has taught us the importance of critical thinking and cross-thinking, and we are grateful for the analytical skills as well as technical skills we manage to gain and apply. We will definitely continue exploring the field, building a version beyond the current Hackathon and bringing some positive changes to the society.
We have learned more about deep learning, in terms of structure and applications. The experiential learning experience enables us to apply what we're taught in school, mentors and workshops provide us with more timely advice on the product and the time constraint pushes us beyond our comfort zone (and procrastination).
Teamwork The most importance accomplishment in the journey is the great teamwork. We literally formed the team 10min before the official start of the hackathon, and are presently surprised that we complement each other in terms of skill sets. As a three-man team, we worked coherently -Yichen did the complex algorithm, Shining built the website and user interface, and Yamei wrote the content. Although our team is smaller, we are proud that we made new friends and built LCD together within the past 2 days.
What we learned
Manage your expectations and stay realistic.
The first lesson is to manage your expectations realistically. Within the short duration, it is nearly impossible to achieve a final design for a tool that normally takes weeks to finish. However, going through the thinking process for the sophisticated algorithm made us optimistic about the future of LCD, and would love to continue working on it. It's good that we tried.
Persistence is the key. There are struggles and challenges, actually much more that we thought. Staying optimistic, making necessary adjustments and never giving up are the critical steps we've learned, not only for hackathons or coding but also for everything we do in life.
Enjoy the process.
Doing things we love and learning new things along the way are simply amazing! The struggles we went through, the pizzas we shared and the product we built eventually are an amazing learning experience and we will do it again.
What's next for Lung Cancer Detection using Deep Learning tool
We are going to upgrade the current product and enhance the accuracy of diagnosis. We will continue research and lab experiments in NYU, consult NYU professors and mentors for future development
Moving on, we will conduct further analysis on the market potential, competitors and industry outlook for the sector. If the growth potential is high, we will consider to commercialize the product and make real changes to people's lives.