Postpartum depression can affect the relationship between mothers and babies. Credit: Kieferpix/iStock/Getty Images Plus
A hybrid deep learning model can be used to identify the initial risk factors for postpartum depression disorder (PPDD)1.
This disorder can lead to severe depression, disrupting the relationship between mothers and babies. An Indian research team says the model can help differentiate women at risk of PPDD from healthy individuals, allowing early medical intervention.
Traditional methods to detect the risk of PPDD rely on survey responses and clinical trial examinations, which can be prone to errors.
To address this issue, scientists at Chandigarh University in Punjab used a hybrid model comprising two separate convolutional neural networks (CNNs) that can decode text and audio data. The model uses Improved Bi-directional Long Short-Term Memory (IBi-LSTM) to process the CNNs' output.
The model was tested on publicly available data from PPDD patients containing 1,550 phrases such as ‘trouble sleeping at night’, ‘irritable towards baby and partner’, ‘overeating and loss of appetite’, ‘suicide attempt’, and ‘feeling sad and tearful’, plus audio data of 195 speech records.
The CNNs decipher main words and phrases related to facial expressions and convert audio data into image-like structures. The IBi-LSTM then combines and processes the CNN-generated features, producing data that show whether or not a woman is depressed.
The model is more precise than other deep-learning methods and shows more than 98% efficiency in identifying women at risk of PPDD. The researchers say it could potentially be used to transmit message alerts to depressed women via mobile apps.
