Our inspiration arrived from existing huge medical databases, with plenty of details, but unmanagable to navigate effectively. We wanted to bring the size and versatility of these databases to the finger tips of everyone around the world, enabling anyone to self-diagnose thier illnesses.
What it does
Diagnose Tri-Testifier is meant to take one of 3 images of skin disease, eye disease, or surface cancer, to have it classified among illness, or danger. Tumors are classified as either Benign or Malignant, while skin and eye diseases are identified.
How we built it
We got information about various illnesses via scientific Databases, identifying Illnesses and their symptoms. We used Azure to scrape the internet of images of each illness, and built corresponding Machine Learning algorithms using Azure. We built front end application that could take, store, and send images, to try to tie the whole project together
Challenges we ran into
As a group, we encountered a huge problem which was scraping the right images from bing to improve the accuracy of our final product. For example; Bing scraped labels of cancer tumour when we wanted symptoms such as spots. So this made it harder for us to improve the precision of our model. So using teamwork, and excellent technical skills we manipulated the code in node.js to search up certain things in a better way so that bing knows specifically what images to take from the internet. ONE BIG issue that we still haven't completed is the fact that we haven't found a way to fulfill a request made by a user using our android app. For example, we want the user to take a picture of their body or syptoms or upload an image from the gallery to the app. We coded these aspects however we were unable to send the request and this image to one of our classifiers. So when the user uploads an image, the image goes through the server with a key to make sure that azure accepts the user then runs it in its designated classifier and returns a message saying what the diseases actually is with an accuracy percentage.We are still figuring which api to use and this was a challenge we yet need to complete.
Accomplishments that we're proud of
We are proud that Diagnose Tri-Testifier Has a working Android interface, along with functional Machine Learning image classification
What we learned
We learned all about Android, Machine Learning and Azure, along with the acuirement of information, we learned about connecting front end to back end, and the integeration of multiple languages