Detection of Alzheimer's Dementia using Large Language Models applied to spontaneous speech

On Monday, October 28, Aula B-221 of the ETSIT UPM (Universidad Politécnica de Madrid) hosted a fascinating talk by Filippo Casu, a former member and friend of the Grupo de Tratamiento de Imágenes (GTI), who is currently doing research at the Università degli Studi di Sassari, Italy.

The presentation, entitled “Detection of Alzheimer's Dementia using Large Language Models applied to spontaneous speech,” attracted an audience interested in his innovative approaches to dementia diagnosis.

Filippo is part of the Ecosystem of Innovation for Next Generation Sardinia project, an initiative aimed at strengthening the connection between science and industry, mitigating the social impacts of the crisis and fostering territorial inclusion. This project seeks to boost innovation, facilitate the transfer of technologies to the productive system, and involve local communities in the challenges of sustainable innovation.

In a region like Sardinia, which has one of the highest elderly population rates in Europe, proactive strategies are being implemented to address the challenges associated with this demographic phenomenon.

Filippo explained that his research focuses on developing improved diagnostic tools for Alzheimer's disease (AD), which affects approximately 50 million people worldwide. Through the use of pre-trained large language models (LLMs), his team achieves automatic detection of AD through linguistic analysis applied to datasets such as ADReSS and ADReSSo (Alzheimer's Dementia Recognition through Spontaneous Speech). This process is performed after converting speech to text, which allows the identification of relevant patterns in patients' spontaneous speech.

While recent advances in LLMs offer a promising basis for their application in healthcare, Filippo stressed the importance of optimizing these models for specific tasks such as AD detection, seeking a balance between performance and computational efficiency. To address data privacy concerns in the healthcare sector, his methodology is implemented on affordable hardware, which facilitates local data processing without compromising security.

Filippo's talk provided attendees with valuable insight into how artificial intelligence and language modeling can revolutionize dementia diagnosis and improve early detection of the disease, thus contributing to a healthier future for older generations.

For more information on Filippo Casu's research: https://doi.org/10.1109/PDP62718.2024.00046