Objective: In this study, we estimate speech features from different Verbal Fluency Tests (VFT) conditions to distinguish comorbid Bipolar Disorder (BD) in adults suffering from Attention Deficit Hyperactivity Disorder (ADHD) and to identify ADHD subtypes such as the inattentive (ADHD-I) from the combined one (ADHD-C). Methods: Prosodic and spectral features in five conditions of VFTs were extracted and selected for the classification performed with machine learning methods. Specifically, a Support Vector Machine exploiting Recursive Features Elimination (SVM-RFE) has been trained with clinical scores and exploiting the within subject variability of speech features across VFT conditions. The final classification was optimized by combining the marginal classification outcomes obtained from the different VFTs using a voting scheme. Results: Our results show that we successfully classify the ADHD+BD comorbidity and the ADHD subtypes according to clinician diagnosis. The results are discussed in the light of possible benefits of developing such approach within clinical research. Conclusion: Significant information is carried out by speech audio features acquired with VFTs, allowing to classify ADHD subtypes and comorbid patterns. This work clearly shows that the audio analysis of speech, along with properly designed speech tasks, is a candidate for the development of clinical decision support systems in psychiatry. Significance: This work represents a major contribution to the applications of speech analysis in ADHD subjects and could support clinicians by identifying objective biomarkers.
Speech signal analysis as an aid to clinical diagnosis and assessment of mental health disorders
Greco A.;Vanello N.
2023-01-01
Abstract
Objective: In this study, we estimate speech features from different Verbal Fluency Tests (VFT) conditions to distinguish comorbid Bipolar Disorder (BD) in adults suffering from Attention Deficit Hyperactivity Disorder (ADHD) and to identify ADHD subtypes such as the inattentive (ADHD-I) from the combined one (ADHD-C). Methods: Prosodic and spectral features in five conditions of VFTs were extracted and selected for the classification performed with machine learning methods. Specifically, a Support Vector Machine exploiting Recursive Features Elimination (SVM-RFE) has been trained with clinical scores and exploiting the within subject variability of speech features across VFT conditions. The final classification was optimized by combining the marginal classification outcomes obtained from the different VFTs using a voting scheme. Results: Our results show that we successfully classify the ADHD+BD comorbidity and the ADHD subtypes according to clinician diagnosis. The results are discussed in the light of possible benefits of developing such approach within clinical research. Conclusion: Significant information is carried out by speech audio features acquired with VFTs, allowing to classify ADHD subtypes and comorbid patterns. This work clearly shows that the audio analysis of speech, along with properly designed speech tasks, is a candidate for the development of clinical decision support systems in psychiatry. Significance: This work represents a major contribution to the applications of speech analysis in ADHD subjects and could support clinicians by identifying objective biomarkers.File | Dimensione | Formato | |
---|---|---|---|
Speech signal analysis as an aid to clinical diagnosis and assessment of mental health disorders.pdf
non disponibili
Tipologia:
Versione finale editoriale
Licenza:
NON PUBBLICO - accesso privato/ristretto
Dimensione
1.75 MB
Formato
Adobe PDF
|
1.75 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
bruno_BSPC.pdf
Open Access dal 01/09/2024
Tipologia:
Documento in Post-print
Licenza:
Creative commons
Dimensione
764.3 kB
Formato
Adobe PDF
|
764.3 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.