Dementia is a progressive neurocognitive disorder affecting over 55 million people globally, a figure projected to rise to 152 million by 2050. Early detection is critical for timely clinical intervention; however, conventional diagnostic methods such as neuroimaging are expensive and mostly inaccessible in low-resource regions. While language impairments provide early cognitive biomarkers, approximately 60% of individuals with dementia are non-English speakers, creating a significant barrier since most speech-based AI frameworks cater primarily to English data. To address this gap, we present an affordable, scalable conversational AI framework utilizing unified multilingual datasets compiled from DementiaBank. By standardizing conversational performance metrics across varied multi-format tasks (including picture descriptions, fluency tests, and narrative tracking), the framework leverages robust multilingual embeddings coupled with binary classifiers to identify language-based biomarkers. Experimental evaluations demonstrate high classification fidelity with strong multilingual generalization, yielding an average diagnostic accuracy of 87% in English and 82% across global languages—offering a highly accessible, cost-effective computational alternative for real-world global clinical screening.

@inproceedings{chennareddy2026dementia,
  title={Affordable Conversational AI For Global Dementia Screening},
  author={Chennareddy, Srihith and Winston, Charisse N},
  booktitle={Alzheimer’s Association International Conference Neuroscience Next},
  year={2026},
  month={February},
  publisher={TalkBank},
  doi={10.5281/zenodo.19043947},
  url={https://doi.org/10.5281/zenodo.19043947}
}