This application, created as part of my academic and professional pursuits in medicine, is a frontend for an AI model developed from academic research for estimating Type 2 Diabetes Mellitus (T2DM) likelihood from voice recordings.
- Accuracy: Optimal accuracy of 75%±22%, specificity of 77%±29%, and sensitivity of 73%±23% in the matched dataset.
- Test Set Performance: 89% accuracy, 91% specificity, and 71% sensitivity.
T2DM seems to create changes to voice amplitude, frequency, harmonic noise, jitter, phonation time, shimmer, and voice turbulence.
If you don't understand the following instructions, give ChatGPT this link and ask it to explain how to run this project:
- Clone the repository.
- Install dependencies:
pip install -r requirements.txt. - Start the Streamlit app:
streamlit run app.py.
Algorithm and methodology adapted from this study.
I've adapted the original code and added a Streamlit interface. This project reflects my passion for merging medicine and technology.
This is a conceptual tool and not a substitute for professional medical advice. Accuracy depends on datasets and conditions similar to those in the original study.
- Adapt core algorithm
- Streamlit frontend
- Testing with more diverse voice samples
- Online community engagement for informal testing
- Local community study in collaboration with healthcare professionals
- Algorithm optimization and improvement
- If there's potential as a screening tool: development of open source smartphone application.
- Planning a formal study for efficacy validation
