Inspiration
Last week I, Saathvik Somujayabalan, was looking at my news feed on my web browser, and this one article caught my eye: Half Of World Lacks Access To Basic Health Care. It also happened to be that my friend, Pratyush Kore, told me about the Hack the World Hackathon. We both wanted to create an application of some sort to at least help address the problem of basic healthcare for all if not solving it. At first, we were both skeptical about registering for this as we knew close to nothing about developing programs or Hackathons in general as neither of us has attended one before. This still didn't bother us and we registered for Hack the World with a clear mission in mind: DigiDoctor for all!
What it does
The user uploads an image of their medical condition that is visible on skin (ex. Acne, hairy cell leukemia) and that image is then fed into our Machine Learning Model (ML). The ML model recognizes the condition and returns to the user its best prediction as to what the condition is.
How we built it
First, we needed to learn all the basics of Machine Learning, Deep Learning, and Neural Networks so we had a foundation to build upon
Then, The data collection for our ML model took up most of our time as we needed a lot of pictures for each of the 16 medical conditions and the 2 control groups (healthy feet, and normal/healthy skin).
We then used the Python libraries TensorFlow and Keras as a platform to build and train our ML model
Finally, we did fine-tunings like adding more data/data augmentation, writing a script to resize all our pictures, and more.
Unfortunately, we only finished the ML model on Friday hence we did not enough time to finish our website to use it on so we turned in our incomplete website and the ML model demo via Command Prompt/Terminal
Challenges we ran into
ML model did not have sufficient data to output a decent accuracy rate (fixed this by implementing data augmentation)
ML model didn't recognize some pictures as they were different sizes (fixed this by writing a script that automatically resized pictures that were fed in)
Accomplishments that we're proud of
We knew absolutely nothing about Machine Learning on Saturday, but now we know exactly how to collect data, train a model, and deploy it!
We were also very proud of ourselves when we tested our ML model for the first time and it correctly predicted the medical condition that we fed in!
What we learned
We learned all the basics of machine learning that was necessary to make DigiDoctor alive
The use of Python APIs to communicate between our Machine Learning Model and website
We also learned how to use Adobe XD to design an appealing website
What's next for DigiDoctor
Integrate this Convolutional Neural Network Model into an app with a GUI to make it easier to use
Add Info pages for each medical condition so when the Machine Learning Model identifies the condition, it will automatically take the user to an Info Page where it will list possible home remedies, recommends if a doctor's attention is necessary, and explain the disease in understandable terms.
Increase ease of access
Expand our list of medical conditions so more can be identified
Constantly adding new and fresh data that our model could use to get better at predicting conditions
Log in or sign up for Devpost to join the conversation.