First Italian-language study combining AI and natural language published in npj Parkinson’s Disease
The Scuola Universitaria Superiore IUSS of Pavia and IRCCS Maugeri Bari are leading a pioneering study that opens new perspectives in the early diagnosis of Parkinson’s disease. Published in npj Parkinson’s Disease (Nature Publishing Group), the study is the first in the world to apply a multivariate model based on artificial intelligence (AI) and natural language processing (NLP) to Italian-speaking patients.
The project is the result of a collaboration between Ailice Labs (IUSS Pavia), the Neuropsychology Laboratory of IRCCS Maugeri Bari, the Global Brain Health Institute (UCSF), Universidad de San Andrés (Argentina), and DeepTrace Technologies, a spin-off of IUSS.
Researchers analyzed speech samples from 40 subjects—with and without Parkinson’s—who performed language tasks such as describing complex images or speaking freely. Advanced machine learning algorithms extracted linguistic variables to train a model capable of distinguishing linguistic traits associated with Parkinson’s.
🔍 Key results:
- 77% accuracy in distinguishing Parkinson’s from healthy controls
- Up to 85% accuracy in classifying cognitive subgroups
- 75% accuracy in identifying cognitive phenotypes (PD-nMCI vs PD-MCI)
“We demonstrated the technical feasibility of analyzing Italian speech,” says Simona Aresta, lead author, biomedical researcher at IRCCS Maugeri and PhD student at IUSS. “It’s a first step toward scalable and remote digital clinical tools.”
Among the most indicative markers: reduced use of action verbs, more frequent speech reformulations, and fewer open-class words (nouns, verbs)—signs of early cognitive decline.“Language offers a unique window into brain function,” explains Dr. Petronilla Battista,neuropsychologist, speech therapist, and Head of the Neuropsychology Lab at IRCCS Maugeri. “Automated speech analysis can support early detection of Parkinson’s cognitive phenotypes and monitor treatment effectiveness.”
“This is the first successful application of a multivariate AI-NLP model to Italian speech in Parkinson’s research,” adds Prof. Christian Salvatore, IUSS faculty member, Director of Ailice Labs, and CEO of DeepTrace Technologies. “Our pipeline is modular, explainable, and ready for clinical integration.”
The research team now aims to:
- expand the study to larger cohorts
- develop explainable and cross-linguistic diagnostic tools
- validate the approach in real-world screening and remote monitoring scenarios
📄 Full reference:
Simona Aresta1, Petronilla Battista1 , Cinzia Palmirotta1, Serena Tagliente1, Gianvito Lagravinese1, Paola Santacesaria1, Allegra Benzini1, Davide Mongelli1, Brigida Minafra1, Christian Lunetta2, Adolfo M. García3,4,5 & Christian Salvatore6,7