Mental Health in Mothers and Babies

Lead Research Organisation: University of Bristol
Department Name: Electrical and Electronic Engineering

Abstract

This project aims to use computational methods to identify behavioural manifestation of problematic parenting in mother-infant interactions. Data modelling techniques will be used to enhance understanding of how mental health problems in a mother are passed on to her child. Such understanding holds the key to breaking an ongoing and reinforcing cycle of mental health disorder risk from generation to generation.

The causes of mental health problems are still poorly understood. However, the association between mother and child mental health disorders is one of the most consistent findings in psychiatry. It is less clear if this is limited to genetic transmission, or if this also translates onto behavioural transmission. This is where novel approaches are critical in order to enhance current understanding. Data modelling in this context will provide critical new evidence regarding the nature of parenting interventions that have potential to improve child mental health and break intergenerational transmission of mental health problems.

The proposed methodology will involve computational modelling of a pre-existing dataset. The dataset comprises a list of mother and child behaviour codes across 15 relevant modalities, taken from video data captured by wearable headcams. Example modalities include alertness, facial expression, vocalisation and touch, and within each mode there are, on average, 6 further detailed options. Additional data is available in terms of the mother's mental health. This data is predominantly continuous, but also involves discrete measures such as mental health test and impulsivity scores.

Initial data analysis will involve computing the frequencies, durations and rates per minute of each individual behaviour. Following this, patterns between modes will be extracted using computational techniques, such as graphical modelling, Bayesian inference and pattern recognition. Additionally, behavioural comparisons will be drawn between mothers with and without mental health difficulties. All programming will most likely take place using Python. This may include the utilisation of pre-existing models/ packages, but will also involve the creation of new ones.

In terms of the PhD structure, this can be roughly broken down by year. Year 1 of the project, I will spend time learning the relevant machine learning (ML) methods that will be applied to the dataset, particularly unsupervised ML concepts such as clustering. This will involve online research and study, but also practicing the newly learned concepts on a smaller subset of the data. Furthermore, I will engage thoroughly will the dataset in question, by spending time learning how the data came to be in it's current format, but also engaging myself in the video coding. Following this, the data will need to be reformatted appropriately for analysis. This will involve exploring which elements of the dataset are important, and which may be removed.

Year 2 will involve initially applying linear analysis to the data, in order to find correlations between modes. From here, graphical models will be implemented in order to further explore the depth of the data. This will involve building sophisticated, non-linear models which are able to combine the many dimensions of the data (i.e. time, behaviour modes, subject). These models will be tested, and the information fed back to the clinicians, parents and psychologists involved. This will be an iterative process, engaging psychologists and data analysts in order to fully scope the depth of the modelling outcomes.

The final project year will involve data visualisation techniques: finding the best and most effective ways of displaying the data, either for the parents or clinicians. This will involve the learning of new concepts, but will follow on well from the previous ML concepts learnt. There will also be considerable write up responsibilities during this year.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023704/1 01/04/2019 30/09/2027
2275429 Studentship EP/S023704/1 23/09/2019 22/09/2023 Romana Burgess