Inspiration
We saw the damage to society caused by depression and wanted to fix this issue. We recognized that many patients could not access depression diagnosis due to a variety of issues. Such as privacy concerns and convenience of travel. We wanted to make a tool that can diagnose depression for everyone regardless of their location or their situation.
What it does
Diagnose depression by analyzing vocal samples from patients.
How we built it
Used deep neural network to train the AI algorithm. We also used tools like triplet learning to improve the accuracy and efficacy of our AI algorithm. We tested our algorithm on a dataset of 53 independent subjects.
Challenges we ran into
Wrong type of encoder which resulted in a low accuracy rate. Wrong selection of feature engineering options which jeopardized the functionality of the algorithm in some initial testing phases.
Accomplishments that we're proud of
95% accuracy when used on a dataset of 53 independent subjects. Teamwork achieved Accurate clustering of data samples on a cluster plot representing actual PHQ-9 scores of that patient.
What we learned
Teamwork and troubleshooting skills. Working under limited amount of time
What's next for Accessipsych, Application for Depression Diagnostic
Real world applications with real patients. Regulation by healthcare governing bodies. International distribution and widespread use.
Built With
- deep-learn
- deep-neural-network
- ml
- triplet-learning
Log in or sign up for Devpost to join the conversation.