The tremulous voice... A confused syllable... The silence extended between words... That cadence pronounced towards the end of the sentence... For any of us, these and other inflections of speech allow us to infer the interlocutor’s intentions and insinuations, to guess his level of fatigue or boredom, and even evaluate your conviction in what you affirm. Now, in expert hands, such phenomena also become key clues to evaluate one of the most prevalent brain conditions in the world: Parkinson’s disease.
Parkinson’s disease affects one in 100 people over the age of 60. It is the second most common neurodegenerative disorder today, only surpassed by Alzheimer’s disease.
Due to neurodegenerative processes that compromise the frontostriatal pathways (brain circuits that connect the basal ganglia with frontal, prefrontal and cerebellar regions), patients manifest motor and cognitive disorders. The former include rigidity, tremors at rest, slowness of movement and, in particular, various forms of dysarthria . The latter have become a focus of growing research interest in multiple fields.
Dysarthria is a set of neuromuscular alterations that affect the ability to control the speech organs, responsible for speech production.
Its many manifestations include anomalies in rhythm (e.g., the relative duration of syllables and rests), intonation (variations in pitch throughout an utterance), and the transition from one sound to another (e.g. ., the ability to activate the vocal cords promptly when going from /p/ to /a/ when saying ’papa’).
The study of dysarthria in Parkinson’s disease is essential for several reasons. 1. First, these deficits affect the intelligibility, self-image and functionality of patients, so that their detection and characterization contribute to clinical and therapeutic tasks. |
However, the evaluation of dysarthria is far from optimal. Typically, subjective weighting is used, in which a single rater, based on his or her impression of the patient’s abilities, establishes scores on five-value scales (where 0 is completely normal and 4 is markedly impaired).
Although this approach is standard worldwide, it has clear limitations, since it depends on highly qualified personnel, tends to show ceiling effects in longitudinal evaluations and, in certain contexts, has low validity and reliability. Furthermore, it is insufficient to establish fine distinctions between the multiple dysarthric profiles documented to date. It happens that, from patient to patient, speech disorders usually vary in a systematic but subtle way.
A group of Latin American researchers faced these challenges, led by doctors Adolfo García (Co-director of the Center for Cognitive Neurosciences at the University of San Andrés, Argentina; and Senior Atlantic Fellow at the Global Brain Health Institute, at the University of California, San Francisco, USA) and Juan Rafael Orozco-Arroyave (Full Professor of the GITA Lab at the Faculty of Engineering of the University of Antioquia, Medellín, Colombia; and Associate Researcher at the Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany).
In an article published in Movement Disorders ( https://doi.org/10.1002/mds.28751 ), García and Orozco-Arroyave’s team implemented a new approach with three distinctive characteristics.
Instead of using subjective assessments, they used automated methods to analyze the acoustic signal of participants’ speech. With these digital technologies, they quantified very diverse and specific aspects at the prosodic (rhythmicity and pitch), articulatory (transition between speech sounds) and phonemic (discriminability of each sound produced) level.
To capture some of the dysarthric variability in the population, they evaluated a group of patients with heterogeneous cognitive profiles and then performed additional analyzes in a subgroup with mild cognitive impairment and another with preserved cognitive abilities, all compared to healthy participants. To do this, they provided two oral production tasks: reading aloud and retelling stories. The data obtained in each one were analyzed using machine learning algorithms, in order to identify the most distinctive dysarthric dimensions of each group.
In the entire group (with cognitive heterogeneity), as well as in the subgroup with preserved cognitive abilities, the highest identification of patients (with accuracy of 84 and 80%, respectively) was achieved by combining prosodic, articulatory and phonemic measures during the reading task .
On the other hand, in the group with mild cognitive impairment, the best patient identification rate was obtained through phonemic analysis during the re-story task (with an accuracy of 87%). This last pattern even allowed us to distinguish between patients from both subgroups with an accuracy greater than 70%. “This finding is notable,” says García, “since the subjective evaluations carried out by expert neurologists fail to discriminate between patients of each subgroup.” Finally, the measures used were able to predict the severity of the patients’ cognitive symptoms.
“These results indicate that automated speech assessments can contribute to the clinical evaluation of Parkinson’s disease, with clear advantages over traditional approaches,” says Dr. Orozco-Arroyave. “A point to highlight – adds García – is that this digital approach overcomes the biases inherent to human weighting, dispenses with specialized clinical personnel, implies a marked reduction in costs and enables massive scalability, even through remote records (e.g. ., through cell phones).
These advantages are particularly relevant in Latin America, where many clinical centers lack specific training options for the evaluation of dysarthria and where the costs of typical approaches are often prohibitive for a large number of patients. In this sense, García concludes, “the strategic use of digital innovations in clinical work represents a commitment to global equity in the field of brain health.”
