Pavia, 22 July 2025 – Artificial intelligence comes to the rescue of the Neurocognitive research and can represent a useful tool for monitoring the effectiveness of pharmacological therapies. The Iuss Higher University School of Pavia and IRCCS Maugeri Bari they sign an important milestone by driving a study that opens new perspectives in the early diagnosis of Parkinson’s disease. Published on the prestigious NPJ Parkinson’s In Disease (Nature Publishing Group) magazine, the work represents the first contribution to the world to apply a multivariate model based on AI and Natural Language Processing (NLP) on Italian language patients. project It was conceived by Maugeri, leader in the rehabilitation and taking charge of fragile patients, jointly with the Iuss of Pavia, a national reference point for advanced research and technological innovation, and aims to Create digital biomarkers of language that allow you to Early identify Parkinson’s disease phenotypes.
Parkinson’s, an algorithm detects falls and alterations of the movement
The study on vocal samples
Table of Contents
The study analyzed i vocal samples collected by 40 IRCCS patients Maugeri Bari Chosen from subjects with diagnosis of Parkinson’s and non -Parkinson’s disease, to whom it was asked to carry out linguistic activities, such as describing complex images or speaking freely, which have been recorded in audio files. The researchers Maugeri and Iuss have processed this data with advanced algorithms, extrapolating linguistic variables (linguistic features) used to train a model of machines learning capable of distinguishing the characteristic features of Parkinson’s patients from those of healthy subjects.
“Promising results”
“We have shown the technical feasibility of analyzing the spoken in Italian language – explains Simona Arstathe first author, a bioengineer researcher at the IRCCS Maugeri in Bari and Doctoranda Iuss at The Hadron Academy -. It is a first step towards digital clinical tools, scalable and applicable even at a distance “.: 77% of accuracy in distinguishing patients with Parkinson’s from healthy subjects; up to 85% in the classification of cognitive subgroups; 75% of performance in distinguishing the two cognitive phenotypes of the disease (PD-NMCI vs Pd-Mci).
Professor Christian Salvatore
The most indicative markers
“Among the most indicative markers that emerged from the analysis, the reduction in the use of action verbs – he underlines Petronilla Battistaneuropsychologist and speech therapist responsible for the neuropsychology laboratory at the IRCCS Maugeri Bari and Corpsontding Author of the study -. These linguistic elements, developed in brain areas such as the frontal lobe, often involved in the early stages of the disease, seem to be particularly sensitive to early deterioration. The project is strategic for Maugeri’s neuromotor rehabilitation medicine department and is one of the innovative technological tools to improve diagnosis and treatment of neurological patients “.
The scenarios
Il Prof. Christian Salvatore, Iuss professordirector of the Ailice Labs Center and CEO of DeepTrace Technologies explains that it is the first time that an approach based on AI and NLP multivariate is successfully applied to distinguish the cognitive profiles of Parkinson’s in Italian: “This work shows how AI can be used to build real digital biomarkers of languagestandardizable, with concrete clinical value. Our pipeline is designed to be modular, explainable, adaptable and integrated into clinical practice for early and non -invasive diagnosis, also applied to other pathological contexts. It is a clear example of translational technology generated in the academic field and ready for impact in clinical practice “.
The next objectives
The next objectives of the research group aim to extend the study to wider clinical championsin order to further consolidate the reliability of the results. In parallel, the development of digital diagnostic tools that are explainable and cross-linguistic, to make them usable in international clinical contexts and guarantee the interpretative transparency of the results is underway. A further step will be the validation of these tools in real scenarios of early screening and remote monitoring, with the aim of offering concrete support to the diagnosis and follow-up of patients.
