-
-
Home Screen of the app
-
Navigation Bar of the application
Inspiration
Alzheimer's disease is an irreversible, progressive brain disorder that slowly destroys memory and thinking skills and, eventually, the ability to carry out the simplest tasks. In 2020, an estimated 5.8 million Americans age 65 and older are living with Alzheimer's dementia. With how detrimental Alzheimer's is, we wanted to build an application that can help healthcare workers and family members identify if a loved one may have Alzheimer's early on.
What it does
This easy to use app allows healthcare workers to submit pictures of an MRI, where our machine learning model can identify if the patient is not demented, very mild demented, mildly demented, or moderate demented. This helps them speed up the process in a busy environment. We then have a Mini-Mental State examination that the user can do in case they do not have an MRI scan, where we use the survey's criteria to determine the patient's level of dementia.
How I built it
We used azure to train a machine learning model, figma to design the application, and flutter to develop it. We used Microsoft Azure Cognition Services which utilizes a technique known as Transfer Learning to predict the severity of Dementia with the precision higher than 98%. In this technique, it trained the last few layers of the model based on our data and parameters and allowed us to export into our application using Tensorflow Lite. We used the dataset from kaggle.com known as Alzheimer's Dataset ( 4 class of Images) which provides us with thousands of images to train and validate our model.
Challenges I ran into
We were able to deploy the app but due to the minute issues we were not able to integrate Firebase and make the mri scan fully functional. Specifically, the image_picker plugin in flutter was much more complicated than we expected it to be. This plugin is used to upload a picture from phone directly to Firebase to scan it using our ml model. But when implementing this plugin we ran into several issues like firebase establishement, storage rules etc.
What I learned
We learned how to use Azure, with it being the first flutter app some of us have built! We have always built our models from scratch in python but it was a great experience building models on cloud platforms through transfer learning and we got to learn about the advantages of cloud based machine learning over the traditional form of machine learning.
What's next for Mental Snap
First and foremost we would like to fix our firebase issues and include all of the features we had initially planned. Secondly, we plan to continually improve the UI to deliver the best user experience. In addition, we plan to use Firebase's SHA storage encryption in the app to secure the user's data and provide a safe platform for everyone.
Log in or sign up for Devpost to join the conversation.