Enhanced AI-Based Methodologies for Detection of Prenatal & Postnatal Depression In Women

Enhanced AI-Based Methodologies for Detection of Prenatal & Postnatal Depression In Women

The introduction of Artificial Intelligence and Machine Learning technologies has been causing a revolutionary change in the field of mental health, especially in prenatal and postpartum depression prediction. It enables healthcare professionals to make timely, informed decisions, which in turn improves mothers’ wellbeing and contribute to family dynamics positively, and improvements in infant development and the mother-infant bond. During delivery (prenatal) and postpartum (1-6 weeks) after childbirth are two of the most critical stages where psychological disturbance remains undiagnosed, which also leads to the main cause of late-stage depression. This thesis investigates, develops, and proposes a triangulation model for prenatal and postnatal mental depression prediction. Organized interviews were used to gather data from women who were admitted for childbirth at SRM Medical College Hospital and Research Centre in Chennai, India. Physiological measures, psychological questionnaire responses, and social media posts make up the dataset. The first model involves an Internet of Things enabled wrist wearable device to monitor the Electro Dermal Activity signals, along with cortisol levels, and Patient Health Questionnaire-9 responses as data sources. Motion artifacts elimination using autoregressive methods, Patient Health Questionnaire-9 responses based data labelling, subject dependent training and independent testing using Leave One Out Cross Validation strategy, an Ensemble Based Deep Learning model was developed to predict the prenatal depression levels and evaluated against benchmark datasets. The second model predicts postnatal or Postpartum depression depression based on psychological questionnaire data (Patient Health Questionnaire-9, Postpartum Depression Screening Scale, and Edinburgh Postnatal Depression Scale). Class imbalance was resolved using data level methods such as data sampling (Over Sampling, Synthetic Minority Over-sampling TEchnique and Under Sampling), and attribute selection (Particle Swarm Optimization). Algorithmlevel methods include MetaCost and ensemble approaches. This hybrid model was evaluated against benchmark datasets and using ablation concepts. The third model identifies postnatal depression markers in social media posts using Attribute Selection Hybrid Network Models model developed with recursive Recurrent Neural Network. Bidirectional Encoder Representations from Transformers attribute extraction algorithm is used for word embedding on social media posts. A vector based analysis with post attention mechanics based on attribute weights are used to predict the Postpartum depression. Further, the trustworthiness of the model was assessed against other benchmark datasets. Thus, the triangulation model improves the prediction and early intervention of maternal depression by Artificial Intelligence based methodologies.

File reproduced in accordance with the copyright policy of the publisher/author/creator.