Search

Introduction: Impaired copy of intersecting pentagons from the Mini-Mental State Examination (MMSE), has been used to assess dementia in Parkinson's disease (PD). We used a digitizing tablet during the pentagon copying test (PCT) as a potential tool for evaluating early cognitive deficits in PD without major cognitive impairment. We also aimed to uncover the neural correlates of the identified parameters using whole-brain magnetic resonance imaging (MRI). Methods: We enrolled 27 patients with PD without major cognitive impairment and 25 age-matched healthy controls (HC). We focused on drawing parameters using a digitizing tablet. Parameters with between-group differences were correlated with cognitive outcomes and were used as covariates in the whole-brain voxel-wise analysis using voxel-based morphometry; familywise error (FWE) threshold p < 0.001. Results: PD patients differed from HC in attention domain z-scores (p < 0.0001). In terms of tablet parameters, the groups differed in Shannon entropy (horizontal in-air, p = 0.003), which quantifies the movements between two strokes. In PD, a correlation was found between the median of Shannon entropy (horizontal in-air) and attention z-scores (R = -0.55, p = 0.006). The VBM revealed an association between our drawing parameter of interest and gray matter (GM) volume variability in the right superior parietal lobe (SPL). Conclusion: Using a digitizing tablet during the PCT, we identified a novel entropy-based parameter that differed between the nondemented PD and HC groups. This in-air parameter correlated with the level of attention and was linked to GM volume variability of the region engaged in spatial attention.

Up to 90% of patients with Parkinson's disease (PD) suffer from hypokinetic dysathria (HD) which is also manifested in the field of phonation. Clinical signs of HD like monoloudness, monopitch or hoarse voice are usually quantified by conventional clinical interpretable features (jitter, shimmer, harmonic-to-noise ratio, etc.). This paper provides large and robust insight into perceptual analysis of 5 Czech vowels of 84 PD patients and proves that despite the clinical inexplicability the perceptual features outperform the conventional ones, especially in terms of discrimination power (classification accuracy ACC = 92 %, sensitivity SEN = 93 %, specificity SPE = 92 %) and partial correlation with clinical scores like UPDRS (Unified Parkinson's disease rating scale), MMSE (Mini-mental state examination) or FOG (Freezing of gait questionnaire), where p < 0.0001., 8 pages, published in International Conference on NONLINEAR SPEECH PROCESSING, NOLISP 2015 jointly organized with the 25th Italian Workshop on Neural Networks, WIRN 2015, held at May 2015, Vietri sul Mare, Salerno, Italy

Automatic objective non-invasive detection of pathological voice based on computerized analysis of acoustic signals can play an important role in early diagnosis, progression tracking and even effective treatment of pathological voices. In search towards such a robust voice pathology detection system we investigated 3 distinct classifiers within supervised learning and anomaly detection paradigms. We conducted a set of experiments using a variety of input data such as raw waveforms, spectrograms, mel-frequency cepstral coefficients (MFCC) and conventional acoustic (dysphonic) features (AF). In comparison with previously published works, this article is the first to utilize combination of 4 different databases comprising normophonic and pathological recordings of sustained phonation of the vowel /a/ unrestricted to a subset of vocal pathologies. Furthermore, to our best knowledge, this article is the first to explore gradient boosted trees and deep learning for this application. The following best classification performances measured by F1 score on dedicated test set were achieved: XGBoost (0.733) using AF and MFCC, DenseNet (0.621) using MFCC, and Isolation Forest (0.610) using AF. Even though these results are of exploratory character, conducted experiments do show promising potential of gradient boosting and deep learning methods to robustly detect voice pathologies., Comment: 11 pages, 1 figure, 10 tables. Keywords: Voice pathology detection, deep learning, gradient boosting, anomaly detection