Ranking Pretrained Speech Embeddings in Parkinson's Disease Detection: Does Wav2Vec 2.0 Outperform its 1.0 Version Across Speech Modes and Languages?

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 31, 2025

Abstract Speech and language technologies are effective tools for identifying the distinct speech changes associated with Parkinson’s disease (PD), enabling earlier more accurate diagnosis. Recent advancements in self-supervised pretraining, particularly Wav2Vec models, have demonstrated superior performance over traditional feature extraction methods. While 2.0 has been successfully utilized PD detection, a rigorous quantitative comparison 1.0 is needed to comprehensively evaluate its advantages, limitations, applicability across different modes PD. This study presents systematic of embeddings three multilingual datasets using various classification approaches classifying normal (healthy controls; HC) speech. Additionally, both versions were benchmarked against baseline features diverse linguistic contexts, including spontaneous speech, non-spontaneous isolated vowels. A multicriteria TOPSIS approach was employed rank methods, revealing that consistently excelled all modes, first transformer layer demonstrating best contextual tasks (read text monologue) extractor performing vowel-based classification. In contrast, 1.0, while generally outperformed by 2.0, still provided faster alternative competitive tasks, highlighting potential specific applications, such as federated learning. comparative analysis furthermore underscores strengths each architecture informs their optimal use detection.

Language: Английский

---

Neuropsychological Tests of Memory, Visuospatial, and Language Function in Parkinson's Disease: Review, Critique, and Recommendations

Movement Disorders, Journal Year: 2025, Volume and Issue: unknown

Published: March 27, 2025

Abstract Background Cognitive impairment in Parkinson's disease (PD) is a key non‐motor complication during the course. Objectives A review of detailed cognitive instruments to detect mild (PD‐MCI) or dementia (PDD) needed establish optimal tests that facilitate diagnostic accuracy. Methods We performed systematic literature assess memory, language including premorbid intelligence, and visuospatial domains (for attention executive functions see accompanying review) determine suitability cognition PD. Based on in‐depth scrutiny psychometric other relevant clinimetric properties, were rated as “recommended,” “recommended with caveats,” “suggested,” “listed” by International Parkinson Movement Disorder Society (IPMDS) panel experts according IPMDS Clinical Outcome Assessment Scientific Evaluation Committee guidelines. Results included 39 encompassing 48 outcome measures. Seven (different versions subtests test counted once) recommended, four for one domains, (including three measures), estimated intelligence. Furthermore, 10 (12 measures) 11 (15 “listed.” Conclusions Recommended neuropsychological functions, are proposed guide assessment its progression PD‐MCI PDD, use clinical trials stratify participants Novel measures being developed will need extensive validation research be “recommended.” © 2025 The Author(s). Disorders published Wiley Periodicals LLC behalf Society.

Language: Английский

---

Investigation of the voice handicaps in Parkinson’s disease and determination of the clinical correlates

Journal of Neural Transmission, Journal Year: 2025, Volume and Issue: unknown

Published: March 24, 2025

Language: Английский

---

Detecting neuropsychiatric fluctuations in Parkinson’s Disease using patients’ own words: the potential of large language models

npj Parkinson s Disease, Journal Year: 2025, Volume and Issue: 11(1)

Published: April 18, 2025

Abstract Over the past decade, neuropsychiatric fluctuations in Parkinson’s disease (PD) have been increasingly recognized for their impact on patients’ quality of life. Speech, a complex function carrying motor, emotional, and cognitive information, offers potential insights into these fluctuations. While previous studies focused acoustic analysis to assess motor speech disorders reliably, linguistic patterns associated with PD remains unexplored. This study analyzed content spontaneous from 33 patients ON OFF medication states, using machine learning large language models (LLMs) predict states state score. The top-performing model, LLM Gemma-2 (9B), achieved 98% accuracy differentiating its predicted scores were highly correlated actual (Spearman’s ρ = 0.81). These methods could provide more comprehensive assessment treatment effects, allowing remote symptom monitoring via mobile devices.

Language: Английский

---

The Newborn Screening Programme Revisited: An Expert Opinion on the Challenges of Rett Syndrome

Genes, Journal Year: 2024, Volume and Issue: 15(12), P. 1570 - 1570

Published: Dec. 5, 2024

Genomic sequencing has the potential to revolutionise newborn screening (NBS) programmes. In 2024, Genomics England began recruit for Generation Study (GS), which uses whole genome (WGS) detect genetic changes in 500 genes more than 200 rare conditions. Ultimately, its purpose is facilitate earlier identification of conditions and thereby improve health-related outcomes individuals. The adoption into GS was guided by four criteria: (1) gene causing condition can be reliably detected; (2) if undiagnosed, would have a serious impact; (3) early or presymptomatic testing substantially outcomes; (4) interventions screened are accessible all. Rett syndrome (RTT, OMIM 312750), paediatric neurodevelopment disorder, not included list GS. this opinion article, we revisit discuss RTT from perspective these criteria. We begin with an introduction then summarise key points about principles, presenting challenges opportunities individuals RTT. provide insight how data could collected during phase, diagnosis our understanding prodromal stage Although many features present departure criteria adopted GS, advances research, combined advocacy parent-based organisations, entry future

Language: Английский

---

Reflexive eye saccadic parameters in Parkinson’s disease

Frontiers in Medical Technology, Journal Year: 2024, Volume and Issue: 6

Published: Nov. 25, 2024

Abnormal reactive saccade, with reduced saccadic gain, impaired smooth pursuit, and unwarranted reactions are clinically used to assess people Parkinson's disease (PwPD). However, there inconsistent findings related other saccade parameters such as latency transition times. This study aimed identify differences in the reflexive of early stage PwPD aged-matched control (AMC).

Language: Английский

---

A methodological framework and exemplar protocol for the collection and analysis of repeated speech samples (Preprint)

Published: Dec. 2, 2024

Speech and language biomarkers have the potential to be regular, objective assessments of symptom severity in several neurological mental health conditions, both in-clinic remotely using mobile devices. However, complex nature speech often subtle changes associated with mean that findings are highly dependent on methodological cohort choices. These not reported adequately studies investigating speech-based assessment, which (i) hinders progress research, (ii) prevents replication, (iii) makes definitive identification robust problematic. 1. To facilitate replicable research by presenting an adaptable collection analytical method design checklist for other researchers adapt their own experiments. 2. develop apply exemplar protocol reduces controls confounding factors repeated recordings healthy speech, including device choice, elicitation task non-pathological variability. We developed a based thematic literature review. Our comprises (1) read (2) held vowels, (3) picture description. With focus towards remote applications, we collected different devices: freestanding condenser microphone, three smartphones headset. report detail pipeline extract set 14 features, also chosen via review, cover timing, prosodic, quality, articulatory, spectral characteristics speech. from 28 individuals times one day, at same 8-11 weeks later, 25 days week fixed times. Participant included sex, age, native status, participant’s voice use habits. Before each recording, information recent use, food drink intake, emotional state. Recording were documented. Values extracted features presented providing resource normative values. data collection, processing, analysis reporting clinical practice varies widely, motivating this checklist. Greater harmonization study protocols consistent urgently required translate processing into practice.

Language: Английский

---

A methodological framework and exemplar protocol for the collection and analysis of repeated speech samples (Preprint)

JMIR Research Protocols, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 2, 2024

Language: Английский

---