Prediction of cerebral palsy in young infants: Computer-based assessment of general movements

Cerebral palsy (CP) is the most serious chronic motor disability that can occur in infants. Despite improvements in care and reduced mortality among high risk term and preterm infants, the prevalence of CP is stable. Diagnosing CP is difficult and early identification of CP might be beneficial for early treatment when the plasticity of the brain is high. Hence, there is a need for accurate assessment methods to provide early clarification about CP development. Assessment of general movements (GMs), a part of the spontaneous movement repertoire, has proven to be reliable and sensitive in the early assessments of infants to identify CP. Using video recordings of infant spontaneous motor activity, observers classify the qualities of general movements into categories. In particular, the absence of so-called fidgety movements in infants at 9-20 weeks post-term age has been shown to be a strong marker for later CP. However, the GM assessment is qualitative and highly dependent on skilled personnel. It is reported to be limited in use in ordinary clinical practice. Recently, computer-vision-based human motion analyses have become possible, providing an inexpensive, non-obtrusive solution for the analysis of movement using video recordings. The aim of this thesis, which constitutes four different scientific studies, was fourfold; 1) to verify the GM assessment for prediction of CP used in a clinical setting and evaluate the inter-rater reliability in the assessment of fidgety movements, 2) to elicit expert knowledge about fidgety movements and identify possible objective characteristics using visual displays, 3) to develop a computer-vision-based method for identification of infants with present and absent fidgety movement characteristics, and 4) to evaluate the accuracy of the developed computer-vision-based method in the prediction of later CP. The results showed that GM assessment by the use of qualitative assessment of fidgety movements demonstrated high sensitivity and specificity in the early prediction of CP. The inter-rater reliability in the assessment of fidgety movements was good. The use of visual displays for observation of absent or present fidgety movements revealed important expert knowledge about periodic movement patterns. The use of a computer-vision-based tool, the General Movement Toolbox, demonstrated ability to detect absent and present fidgety movement characteristics and high sensitivity and specificity for prediction of CP during the fidgety movement period. The General Movement Toolbox also demonstrated high sensitivity and specificity, correctly predicting ambulatory or non-ambulatory function for 9 of 10 children with CP. This thesis provides a novel computer-based method for early prediction of CP in young infants based on a single video recording. The assessment instrument General Movement Toolbox is presented, showing promising prediction of CP and ambulatory versus nonambulatory function in children with CP. This thesis adds an early objective detection of movement hallmarks for later neurological disease to the field of neuromotor assessment of infants. It brings hope for earlier and more precise prediction of CP and evaluation on early intervention strategies in future research