Periodic Reporting for period 1 - MULTI-LAND (Multicentric Language Markers of NeuroDegeneration)
Période du rapport: 2021-09-01 au 2023-08-31
In the next decade, Europe will experience a significant demographic shift, with over 25% of its population surpassing the age of 65. This foreshadows a dramatic growth of aging-related neurodegenerative diseases (NDs). Among the most prevalent NDs, impacting approximately 50 million individuals, are Alzheimer's disease (AD), the behavioural variant of frontotemporal dementia (bvFTD), and Parkinson's disease (PD). Despite shared neuropathological features, these diseases exhibit distinct cognitive and brain connectivity profiles. AD primarily affects temporo-parieto-hippocampal networks and memory function, bvFTD targets fronto-insulo-temporal connectivity and is associated with personality and social cognition deterioration, while PD impacts fronto-striatal networks and motor cognitive skills. Interestingly, these NDs are associated with distinct language deficits, with AD undermining lexico-semantic skills, bvFTD disrupting socio-emotional concepts, and PD affecting action verbs and morpho-syntactic aspects.
Currently, diagnosing and monitoring NDs rely on lengthy, stressful cognitive tests and costly brain scans, which can be particularly challenging in regions with economic disparities. To overcome these limitations, the MULTI-LAND project aims to implement a ground-breaking approach centred on natural language markers (NLMs) extracted from patients' speech and analysed via machine learning algorithms. Although NLMs have shown promise in detecting mental health conditions, their potential application in NDs remains largely unexplored. Furthermore, no study has yet (i) investigated the association of these language markers with brain network atrophy and oscillatory aberrations, (ii) assessed their robustness, or (iii) evaluated their diagnostic potential and generalizability across different research centres and countries.
The primary research objective of MULTI-LAND is to conduct a comprehensive, cross-methodological (behavioural, MRI/fMRI, EEG) and multi-center (i.e. the BCBL in Europe and the CNC-UdeSA in Argentina) validation of NLMs. The overarching hypothesis is that NLMs will be associated with disease-specific, cognitive and neural patterns across NDs, offering a cost-effective, scalable, and remotely applicable approach for early ND detection
Currently, diagnosing and monitoring NDs rely on lengthy, stressful cognitive tests and costly brain scans, which can be particularly challenging in regions with economic disparities. To overcome these limitations, the MULTI-LAND project aims to implement a ground-breaking approach centred on natural language markers (NLMs) extracted from patients' speech and analysed via machine learning algorithms. Although NLMs have shown promise in detecting mental health conditions, their potential application in NDs remains largely unexplored. Furthermore, no study has yet (i) investigated the association of these language markers with brain network atrophy and oscillatory aberrations, (ii) assessed their robustness, or (iii) evaluated their diagnostic potential and generalizability across different research centres and countries.
The primary research objective of MULTI-LAND is to conduct a comprehensive, cross-methodological (behavioural, MRI/fMRI, EEG) and multi-center (i.e. the BCBL in Europe and the CNC-UdeSA in Argentina) validation of NLMs. The overarching hypothesis is that NLMs will be associated with disease-specific, cognitive and neural patterns across NDs, offering a cost-effective, scalable, and remotely applicable approach for early ND detection
The major accomplishment during this project phase is the successful identification of natural language markers that distinguish AD patients from those with bvFTD and healthy controls. All participants engaged in a series of fluency tasks, and for the analysis of their speech responses, we employed two distinct approaches. Firstly, we adopted the classical method, which involves counting valid responses. Secondly, we introduced a novel approach, in which we computed several linguistic properties (e.g. frequency, granularity, neighbourhood, length, familiarity, and imageability) for each spoken word. These properties were used for group-level differentiation, patient-level identification, and for assessing their association with cognitive metrics (e.g. executive control) and neural patterns (f/MRI and EEG).
While standard scores, based on valid responses, did reveal deficits in both patient groups, our word-property approach uncovered specific alterations in AD, particularly in the domains of frequency, granularity, and phonological neighbourhood. Specifically, we find that, when navigating their semantic memory, individuals with AD showed a strong preference for producing easily accessible words, characterized by high usage, conceptual unspecificity (e.g. flower instead of rose), and similar phoneme sequences (i.e. such as rat relative to cat).
In parallel, using machine learning techniques, we found that these word-properties allowed robust subject-level classification performance (AUC = 0.89 ± 0.09) exclusively for individuals with AD. Furthermore, these linguistic attributes also predicted executive function outcomes only in AD patients. When considering brain measures, AD patients showed reduced cortical thickness in temporo-parietal areas and hypoconnectivity in critical resting-state fMRI networks (e.g. default mode network) and in both cases these patterns correlated with word frequency. Lastly, EEG functional hypoconnectivity within the AD group correlated with frequency, granularity, and phonological neighbourhood, underscoring the link between neural patterns and linguistic profiles
While standard scores, based on valid responses, did reveal deficits in both patient groups, our word-property approach uncovered specific alterations in AD, particularly in the domains of frequency, granularity, and phonological neighbourhood. Specifically, we find that, when navigating their semantic memory, individuals with AD showed a strong preference for producing easily accessible words, characterized by high usage, conceptual unspecificity (e.g. flower instead of rose), and similar phoneme sequences (i.e. such as rat relative to cat).
In parallel, using machine learning techniques, we found that these word-properties allowed robust subject-level classification performance (AUC = 0.89 ± 0.09) exclusively for individuals with AD. Furthermore, these linguistic attributes also predicted executive function outcomes only in AD patients. When considering brain measures, AD patients showed reduced cortical thickness in temporo-parietal areas and hypoconnectivity in critical resting-state fMRI networks (e.g. default mode network) and in both cases these patterns correlated with word frequency. Lastly, EEG functional hypoconnectivity within the AD group correlated with frequency, granularity, and phonological neighbourhood, underscoring the link between neural patterns and linguistic profiles
Our research goes beyond the current state of the art by unveiling speech markers specifically tailored for Alzheimer's disease. Our findings show the presence of unique word-property characteristics in AD, absent in other forms of dementia. These characteristics allow for better identification of AD at both group and individual levels, while also capturing essential cognitive and brain signatures. Crucially, our approach is objective, automated, and low-cost, making it a promising candidate for integration into user-friendly clinical applications.
The socio-economic and broader societal implications of our work are thus substantial. Most notably, the incorporation of these language markers into clinical assessments holds the potential to enhance the accuracy and efficiency of neurodegenerative disease diagnosis. This is especially critical in the context of a global aging population, ensuring that timely interventions and personalized care can be provided to patients. Furthermore, our approach's cost-effectiveness and affordability can help to alleviate the financial burden on healthcare systems, especially in economically disadvantaged regions.
The next phase of the project focuses on validating this set of natural speech markers in European populations. This step is essential to ensure the reliability of our findings across diverse cultural and linguistic contexts.
The socio-economic and broader societal implications of our work are thus substantial. Most notably, the incorporation of these language markers into clinical assessments holds the potential to enhance the accuracy and efficiency of neurodegenerative disease diagnosis. This is especially critical in the context of a global aging population, ensuring that timely interventions and personalized care can be provided to patients. Furthermore, our approach's cost-effectiveness and affordability can help to alleviate the financial burden on healthcare systems, especially in economically disadvantaged regions.
The next phase of the project focuses on validating this set of natural speech markers in European populations. This step is essential to ensure the reliability of our findings across diverse cultural and linguistic contexts.