Machine learning to understand the impact of epilepsy on functional outcomes

This exciting multi-disciplinary PhD project will develop machine learning approaches to understand the impact of epilepsy on functional outcomes of patients.  The successful applicant will do research in a multi-disciplinary and cross-College setting at the interface of machine learning, clinical practice, and neurology.  This PhD provides an excellent opportunity to be trained in quantitative and interdisciplinary skills.

Background

Epilepsy is a complex disease with devastating effects on quality of life of children, young people, and their families.  Pediatric epilepsy and functional impairments frequently coexist and further affect quality of life.  Timely identification of these problems is critical for early interventions and the provision of support to patients and their families.  This urgent need for early recognition of impairment is highlighted in calls to action by bodies including the International League Against Epilepsy and World Health Organization^[1].  The gold standard to identify this kind of problems is detailed neuro-developmental assessments.  These may not be readily available, are time-consuming, and can be clinically challenging.

We urgently need objective, reliable, personalised, non-invasive markers beyond current standard approaches to identify patients at high risk of functional problems (or mental disorders).

Children with epilepsy in Scotland have routine Magnetic resonance imaging (MRI) of brain and electroencephalogram (EEG) performed during disease diagnosis and subsequent monitoring.  What if we could take advantage of this available routinely collected clinical data and use a data-driven technique to assess the children’s risk of impairments? To-date, these routinely collected radiological and neurophysiological data remain an under-used resources for objective, personalised functional and structural assessments.  Collaborations between clinicians and data scientists would allow application of algorithms on these under-investigated modalities to develop a clinically useful model in objective outcome assessment of children with epilepsy^[2].  This may in turn help clinicians to improve treatment and their patients’ quality of life.

Studies have shown that EEG connectivity is altered prior to seizures and at baseline, for example, in children with developmental comorbidities such as autism.  Changes in neural network connectivity, as measured by phase synchrony, provide objective markers of brain function.  The pattern of network synchronization or desynchronization is unique to the individual patient^[3]. Furthermore, there is evidence of correlation between structural MRI and behavioural scores in children with epilepsy.  This correlation agrees with previous cross-sectional studies of children with chronic localization-related epilepsy using traditional MRI volume analysis, which revealed distributed changes in volumes across a range of regions, including subcortical structures^[4].

Aims

The successful applicant will address the following questions:

• Can EEG connectivity calculation be used as an objective outcome assessment for behavioural problems in patients with epilepsy?
• Does EEG connectivity outcome assessment correlate to structural outcome assessment on brain MRI in these patients?

For this, they will develop and apply data science and machine learning algorithms to predict developmental scores in children with early onset epilepsy by combining EEG, sMRI, and phenotypic data (Strength and Disability Questionnaires) through advanced machine learning methods.

Training outcomes

We provide a unique opportunity for the successful applicant to be trained in a range of skills, including digital, quantitative, clinical, and interdisciplinary ones, and to apply these skills to a variety of clinical data types and sources.

References

[1] Brooks-Kayal, “Issues related to symptomatic and disease-modifying treatments affecting cognitive and neuropsychiatric comorbidities of epilepsy”, Epilepsia (2013).

[2] Dron, “Functional, structural, and phenotypic data fusion to predict developmental scores of pre-school children based on Canonical Polyadic Decomposition”, Biomed Signal Process Control (2021).

[3] Barstein, “Resting state EEG abnormalities in autism spectrum disorders”, J Neurodevelop Disord (2013).

[4] Yoong, “Quantifying the deficit–imaging neurobehavioural impairment in childhood epilepsy”, Quantitative Imag. Med. Surg. (2015).