ECE2024 Oral Communications Oral Communications 5: Pituitary and Neuroendocrinology | Part I (6 abstracts)
Digital voice analysis as a biomarker for acromegaly
Konstantina Vouzouneraki 1 , Fredrik Nylén 2 , Jenny Holmberg 2 , Tommy Olsson 3 , Katarina Berinder 4,5 , Charlotte Höybye 4,5 , Sophie Bensing 4,5 , Maria Petersson 4,5 , Anna-Karin Ækerman 6,7 , Henrik Borg 8,9 , Bertil Ekman 10,11 , Jonas Robért 10 , Britt Edén Engström 12,13 , Oskar Ragnarsson 14,15,16 , Pia Burman 8,9 & Per Dahlqvist 1
1Umeå University, Department of Public Health and Clinical Medicine, Umeå, Sweden; 2Umeå University, Department of Clinical Sciences, Umeå, Sweden; 3Umeå Unversitty, Department of Public Health and Clinical Medicine, Umeå, Sweden; 4Karolinska Institutet, Department of Molecular Medicine and Surgery, Stockholm, Sweden; 5Karolinska University Hospital, Department of Endocrinology, Stockholm, Sweden; 6Örebro University Hospital, Department of Internal Medicine, Örebro; 7Örebro University, Faculty of Health and Medical Sciences, Örebro, Sweden; 8Lund University, Department of Endocrinology, Malmö, Sweden; 9Skåne University Hospital, Department of Endocrinology, Malmö, Sweden; 10Linköping University, Department of Health, Medicine and Caring Sciences, Linköping, Sweden; 11Linköping University Hospital, Departments of Endocrinology in Linköping and Norrköping, Linköping, Sweden; 12Uppsala University, Department of Medical Sciences, Endocrinology and Mineral Metabolism, Uppsala, Sweden; 13Uppsala University hospital, Department of Medical Sciences, Endocrinology and Mineral Metabolism, Uppsala, Sweden; 14Institute of Medicine at Sahlgrenska Academy, University of Gothenburg, Department of Internal Medicine and Clinical Nutrition, Gothenburg, Sweden; 15Sahlgrenska University Hospital, Department of Endocrinology, Gothenburg, Sweden; 16Wallenberg Centre for Molecular and Translational Medicine, Institute of Medicine, Gothenburg, Sweden
Background: Due to the rarity and the slow progress of the disease, there is a considerable diagnostic delay in acromegaly, contributing to increased morbidity and mortality. Voice analysis may decrease this delay as patients with acromegaly often develop a dark voice and unclear speech due to macroglossia, enlarged lips and edema in the pharynx and larynx. Modern technology offers higher speed and power of voice analysis and increased the interest in voice as a biomarker to identify several diseases. The aim of this study was to develop an algorithm identifying patients with acromegaly using a wide range of acoustic parameters and machine learning.
Method: Voice recordings from patients with acromegaly and from matched controls were collected using a mobile phone with a sound recording application (Voice Record Pro) at all University hospitals in Sweden. The recordings included a sustained “aaaaa”. Anthropometric and clinical data and self-reported voice handicap index (VHI) were assessed. The voice analysis of the sustained “aaaaa” included 3274 parameters, used for training of three different models (Support Vector Machines, Random forests and k-Nearest Neighbor) to classify the speaker as “acromegaly” or “control”. The models were combined into a model ensemble, a machine learning approach to combine models in the prediction process to increase the accuracy of classification. The model was trained with 75% of available data and its performance was assessed using the remaining 25% of the data set.
Results: We included 151 (39% women) Swedish patients with acromegaly, 77% in biochemical remission, 10% biochemically active despite treatment and 13% recently diagnosed and treatment naive. For comparison 139 matched controls (41% women) were included. The model ensemble could identify patients with acromegaly with a sensitivity of 71% and a specificity of 77% (diagnostic accuracy 74%, ROC AUC 84%) based on voice classification of the sustained “aaaaa”. Self-reported voice problems (VHI) were significantly higher in patients with acromegaly compared to controls (median VHI 6 vs 2, P<0.01). The proportion of individuals with VHI >20, representing a clinically significant voice impairment was higher in the acromegaly group than the control group (22.5% vs 3.6%).
Conclusions: Modern voice analysis can identify patients with acromegaly from short voice recordings with a reasonable accuracy. Patients with acromegaly have an increased prevalence of self-reported speech and voice disturbances even after biochemical control of acromegaly.
Volume 99
Article tools
My recent searches
My recently viewed abstracts
Authors
Endocrine Abstracts
ISSN 1470-3947 (print) | ISSN 1479-6848 (online)
© Bioscientifica 2024 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more