Automated injury segmentation to assist in the treatment of children with cerebral palsy

Abstract

Cerebral palsy (CP) describes a group of permanent disorders of posture and movement caused by disturbances in the developing brain. It is the most common physical disability of children worldwide and can lead to a wide range of functional impairments. Accurate diagnosis and prognosis, in terms of the type and severity of functional impairment, is difficult due to the wide range of injuries that may occur and the variable effect of plasticity; which leads to inconsistency in the clinical outcomes of children with CP. The use of Magnetic resonance imaging (MRI) to identify and locate brain lesions has facilitated the diagnosis and qualitative classification of children with CP.Currently, the quantification of image findings from structural MRIs is not automated and remains labour intensive, hence is not widely performed for clinical assessment. Automated brain image segmentation techniques could reduce the clinical time required toprovide an accurate and reproducible quantification of injury. Although such approaches have been used to study other neurological disorders, the heterogeneous appearance of injury and the large anatomical distortion that occurs in CP require modification of existing algorithms to be sufficiently reliable. As a result, they have not been widely applied toMRIs of children with CP.In this thesis, a series of automated image quantification techniques are presented for analysing the MRI data obtained from children with CP. These techniques identify and quantify the severity of the three main types of injury observed in children with CP; including ventricular enlargement, cortical malformations and white and grey matter injury.The developed automated pipeline involves a number of technical developments and contributions to the automated analysis of CP MRI data. A brain tissue segmentation approach based on the Expectation Maximisation (EM)/Markov Random Field (MRF) approach was developed, with a modified MRF implementation that expects different tissue labels within a given neighbourhood to have a corresponding intensity gradient.Following this segmentation step, three different approaches were used to identify and quantify the three distinct types of injury observed from children with CP, which are all important for clinical assessment. Biomarkers from each type of injury obtained from these approaches were used as independent variables in a devised statistical methodology designed to elucidate significant and generalisable correlations between image-derived measures of injury and patient function.ii Ventricular enlargement was quantified using a model of lateral ventricular shape that encapsulates healthy variation observed in ventricular shape. The residual of this model to a target shape reveals a volume of injury, allowing a measure of involvement of critical adjacent anatomies, such as the thalamus, caudate nucleus and lenticular nucleus, to be computed. This measure of involvement was devised as a way to translate the indirect injury of ventricular enlargement to the direct in...

Individual Variation of Human Cortical Structure Is Established in the First Year of Life

“…For example, cortical gray matter (GM) volumes, cortical thickness (CT) and surface area (SA) are all reduced in schizophrenia (1,2); these abnormalities are present in the early stages of clinical illness and in children at risk (3)(4)(5). Autism is also associated with alterations of cortical GM structure, including increased CT and SA (6)(7)(8). Cortical GM abnormalites are present in attention deficit-hyperactivity disorder (9), bipolar disorder (10), and conduct disorders (11) and major depression (12).…”