Assessing Mothers’ Postpartum Depression From Their Infants’ Cry Vocalizations

Abstract

Postpartum Depression (PPD), a condition that affects up to 15% of mothers in high-income countries, reduces attention to the needs of the child and is among the first causes of infanticide. PPD is usually identified using self-report measures and therefore it is possible that mothers are unwilling to report PPD because of a social desirability bias. Previous studies have highlighted the presence of significant differences in the acoustical properties of the vocalizations of infants of depressed and healthy mothers, suggesting that the mothers’ behavior can induce changes in infants’ vocalizations. In this study, cry episodes of infants (N = 56, 157.4 days ± 8.5, 62% firstborn) of depressed (N = 29) and non-depressed (N = 27) mothers (mean age = 31.1 years ± 3.9) are analyzed to investigate the possibility that a cloud-based machine learning model can identify PPD in mothers from the acoustical properties of their infants’ vocalizations. Acoustic features (fundamental frequency, first four formants, and intensity) are first extracted from recordings of crying infants, then cloud-based artificial intelligence models are employed to identify maternal depression versus non-depression from estimated features. The trained model shows that commonly adopted acoustical features can be successfully used to identify postpartum depressed mothers with high accuracy (89.5%).

A Machine Learning Approach for the Automatic Estimation of Fixation-Time Data Signals’ Quality

“…The technique consisted of generating new synthetic samples by adding white noise to a copy of the whole or of a subset of the original sample [39,44,45]. The technique has been shown to be suitable for ML analysis and classification of different type of signals, including neurophysiological signals [27,46], and it has been successfully employed to enhance the accuracy of ML classifiers [39,47].…”

“…Previous works have shown that Machine Learning (ML) models can be successfully employed to study neurophysiological signals. In Gabrieli et al [27], for example, different machine learning models were tested to verify the possibility of classifying infants' vocalisations while in other works the technique was shown to be suitable for the automatic identification of physiological signal quality [28,29]. Li et al [29], for example, employed Support Vector Machines (SVM) models to distinguish between clean and noisy electrocardiogram (ECG) recording.…”

Evaluation of the Severity of Major Depression Using a Voice Index for Emotional Arousal

“…Globally, economic loss due to mental health disorders is an issue that requires immediate and appropriate attention [ 1 , 2 ]. In addition, early detection of depression can help to reduce the number of suicides caused by depression and prevent infanticide by mothers with postpartum depression (PPD) [ 3 ]. To address these issues, an easily accessible and low-cost mental health screening method is required.…”

“…Reporting bias refers to participants’ intentional disclosure or suppression of certain information (e.g., medical history, smoking history). PPD is usually identified using self-report measures; therefore, it is possible that mothers with PPD are unwilling to report it because of social desirability bias [ 3 ]. Further, assessment tools using biomarkers such as saliva and blood have also been studied [ 8 , 9 , 10 , 11 ].…”

“…Furthermore, Gabrieli et al [ 3 ] investigated whether there is a significant difference in the acoustical properties of the vocalizations of infants between depressed and healthy mothers. They extracted acoustic features (fundamental frequency, first four formants, and intensity) from recordings of crying infants.…”