Our main inspiration for this venture comes from the movie 'Love and Other Drugs', where we witness the plight of the people as they struggle with Parkinson's, including the lead actress, played by Anne Hathaway. The aim behind this application is to aid in the diagnosis process and to make sure that we don't lose someone as lovely as Anne Hathaway to Parkinson's.
What it does
The system takes in speech and sensor input from the user implicitly. We currently use the accelerometer sensor data to be able to detect and diagnose with tremors in the hands. Speech input is recorded by interacting with our very own virtual helper, Anne, who interacts with the user as feedback is being recorded. This collected data is run on a predictive machine learning model, which detects the likelihood of a person being diagnosed with Parkinson's.
How we built it
We used the Django framework in Python to build the overall application. Speech data and sensor data were collected through the application interface and a trained neural network was used to make predictions on this data. The neural network was trained on the data collected by a study in a lab at Oxford, which was available as open-source data. Our virtual helper was designed using the SitePal tool.
Challenges we ran into
One of the major challenges that we ran into while developing the application was in developing the Virtual Helper as most of our team has little to no experience in building VR applications. Another challenge we faced was in reducing the delay in training the model. We dealt with this challenge by considering the training step as a pre-requisite step in the design, so that the model is trained long before data is collected from the user.
Accomplishments that we're proud of
Considering that our team had no experience in developing VR applications, our attempt at building a virtual helper for the application is certainly something to be proud of. We are also particularly fond of our idea, as we discovered that there has not been a lot of research in using Machine Learning to help in diagnosis of Parkinson's.
What we learned
We learned a lot about developing an end-to-end Machine Learning application, right from the collection of data, preprocessing and design considerations on delivering results efficiently. We also learned about how Machine learning can be used to build really useful systems that can be used to help the society at large.
What's next for Detection Of Parkinson's Disease Using Neural Networks
Future work for our application in specific would be to improve on the VR to enhance the overall user experience, improve the efficiency of our Machine learning model by using more powerful infrastructure such as AWS's P1 and P2 instances and employing more powerful data processing techniques such as SMOTE for balancing data bias.