SERPICO 2020 Abstracts


Full Papers
Paper Nr: 1
Title:

Parkinson’s Disease Glottal Flow Characterization: Phonation Features vs Amplitude Distributions

Authors:

Agustín Álvarez, Andrés Gómez, Daniel Palacios, Jiri Mekyska, Athanasios Tsanas, Pedro Gómez and Rafael Martínez

Abstract: The study of speech and voice in people diagnosed with a neurodegenerative disorder for the purposes of detection and monitoring has known a very relevant push forward in these last years, but it is far from being completed. One of the main concerns nowadays is that once the deterioration of speech and phonation quality has been informed by machine learning relying upon clinical expertise, there is insufficient evidence to resolve if quality deterioration may come from organic causes, neuromotor degeneration or simply from aging. The present work is part of a more ambitious plan to shed light on this problem by resorting to a theoretical modelling of glottal signals under the main known causes affecting phonation quality, which are closure deficits during the phonation cycle. These deficits may be due to anatomical, organic pathologic or neuromotor reasons. Simulation examples explaining them in the glottal excitation signals are given and contrasted with real examples. Finally, relevant scores from an experimental separation of Parkinson Disease phonation samples from 24 male and 24 female subjects against aging 24 male and 24 female controls on the same age taken from a male-female balanced dataset confronted to a normative subset of 24 male and 24 female speakers are presented to exemplify an analysis study deepening into this problem. Although classification accuracy scores as high as 99.69 and 99.59 were attained in 10-fold cross-validation using an SVM classifier, there is still the impression that co-morbidity and aging effects are not well taken into account, requiring a further semantic study on the features behind the discrimination scores obtained.

Paper Nr: 2
Title:

Large-scale Clustering of People Diagnosed with Parkinson’s Disease using Acoustic Analysis of Sustained Vowels: Findings in the Parkinson’s Voice Initiative Study

Authors:

Athanasios Tsanas and Siddharth Arora

Abstract: The heterogeneity of symptoms in Parkinson’s Disease (PD) has motivated investigating PD subtypes using cluster analysis techniques. Previous studies investigating PD clustering have typically focused on symptoms assessed using standardized clinical evaluations and patient reported outcome measures. Here, we explore PD subtype delineation using speech signals. We used data from the recently concluded Parkinson’s Voice Initiative (PVI) study where sustained vowels were solicited and collected under non-controlled acoustic conditions. We acoustically characterized 2097 sustained vowel /a/ recordings from 1138 PD participants using 307 dysphonia measures which had previously been successfully used in applications including differentiating healthy controls from PD participants, and matching speech dysphonia to the standard PD clinical metric quantifying symptom severity. We applied unsupervised feature selection to obtain a concise subset of the originally computed dysphonia measures and explored hierarchical clustering combined with 2D-data projections using t-distributed stochastic neighbor embedding to facilitate visual exploration of PD subgroups. We computed four main clusters which provide tentative insights into different dominating speech-associated pathologies. Collectively, these findings provide new insights into the nature of PD towards exploring speech-PD data-driven subtyping.

Paper Nr: 3
Title:

Assessing Preferred Proximity Between Different Types of Embryonic Stem Cells

Authors:

Minhong Wang, Athanasios Tsanas, Guillaume Blin and Dave Robertson

Abstract: Embryonic stem cells (ESCs) studies play an important role for understanding the molecular events that underlie cell lineage commitment and serve as models for the development of disease. However, the interactions between neighboring embryonic stem cells are not fully understood. Assessing proximity between different types of embryonic stem cells might provide more information about distinct behaviors of embryonic stem cells. In this study, we processed 186 cell colonies on disc constrained microdomains and 152 cell colonies on ellipse. We grouped cell colonies based on different observed patterns and grouped cells by their locations. By applying two measurements on embryonic stem cell colonies, minimum spanning tree and average distance to the five closest objects, we investigated the difference of proximity between different types of embryonic stem cells, the difference between grouped cell colonies and the difference between grouped cells. We found one type of ESC has a smaller average path based on minimum spanning tree and higher proximity than the other type. We report consistent results for different types of embryonic stem cells: these findings may be useful to set benchmarks for empirical models which replicate ESC behaviors.

Paper Nr: 4
Title:

Data-Driven Insights towards Risk Assessment of Postpartum Depression

Authors:

Evdoxia Valavani, Dimitrios Doudesis, Ioannis Kourtesis, Richard M. Chin, Donald J. MacIntyre, Sue Fletcher-Watson, James P. Boardman and Athanasios Tsanas

Abstract: Postpartum depression is defined as depressive episodes that occur during pregnancy or within 12 months of parturition. The goal of this study is the exploration of the birth features and maternal traits which affect the risk of postpartum depression for mothers with preterm neonates. We analysed data from 144 women (63 mothers of term and 81 mothers of preterm infants) who completed the Edinburgh Postnatal Depression Scale (EPDS) in the postpartum period. We used three feature selection algorithms: ReliefF, Random Forests (RF) variable importance, and Boruta, in order to select the most predictive feature subsets, which were subsequently mapped onto the binarized EPDS total score (a threshold of 10 was used to binarize the EPDS total scores) using RF. We found that positive affectivity (rs=-0.467, p<0.001), and the Apgar score at 5 minutes (rs=-0.430, p<0.001) are the most statistically strongly associated features with the risk of postpartum depression. We used 10-fold cross-validation with 100 iterations and report out-of-sample balanced accuracy (median±IQR): 75.0±16.7, sensitivity: 66.7±16.7, specificity: 100±16.7, and F1 score: 0.8±0.2. Collectively, these findings highlight the potential of using a data-driven process to automate risk prediction using standard clinical characteristics and motivate the deployment of the developed tool using larger-scale datasets.