Project

dementAI

Utkarsh Sarawgi

 dementAI is an open-source platform for modeling risk stratification of Alzheimer's dementia using spontaneous speech through a mobile service powered by privacy-protected and interpretable AI.

Alzheimer’s disease is estimated to affect around 50 million people worldwide and is rising rapidly, with a global economic burden of nearly a trillion dollars.  We are developing a scalable, cost-effective, and robust method for stratification of Alzheimer’s dementia using a language-agnostic mobile service. Our approach uses a multi-modal ensemble of acoustic and cognitive markers (like prosody, disfluency, tracking user's train of thoughts, and more) derived from the user’s spontaneous speech.

Our uncertainty-aware and multi-modal AI system has achieved state-of-the-art results on benchmark dementia datasets, and shows the promise of mobile and non-invasive applications for risk stratifying conditions like dementia. The size and minimal design of our models allow for on-device training and inference, thus supporting data privacy. We baked interpretability into our system design to help healthcare providers with … View full description

 dementAI is an open-source platform for modeling risk stratification of Alzheimer's dementia using spontaneous speech through a mobile service powered by privacy-protected and interpretable AI.

Alzheimer’s disease is estimated to affect around 50 million people worldwide and is rising rapidly, with a global economic burden of nearly a trillion dollars.  We are developing a scalable, cost-effective, and robust method for stratification of Alzheimer’s dementia using a language-agnostic mobile service. Our approach uses a multi-modal ensemble of acoustic and cognitive markers (like prosody, disfluency, tracking user's train of thoughts, and more) derived from the user’s spontaneous speech.

Our uncertainty-aware and multi-modal AI system has achieved state-of-the-art results on benchmark dementia datasets, and shows the promise of mobile and non-invasive applications for risk stratifying conditions like dementia. The size and minimal design of our models allow for on-device training and inference, thus supporting data privacy. We baked interpretability into our system design to help healthcare providers with insights into dementia. The modularity of our speech platform may further support other AI-assisted speech applications such as affect recognition, monitoring depression, and more. The project's  codebase is entirely open-source to promote research around scalable, cost-effective, and robust methods using spontaneous speech for dementia and other conditions.

For further information, see our paper "Multimodal Inductive Transfer Learning for Detection of Alzheimer’s Dementia and its Severity" accepted to the INTERSPEECH Conference, 2020.  Two additional papers "Why have a Unified Predictive Uncertainty? Disentangling it using Deep Split Ensembles" and "Uncertainty-Aware Multi-Modal Ensembling for Severity Prediction of Alzheimer's Dementia" are currently under review. 

Project at a glance

Person People
Utkarsh Sarawgi
Research Assistant