Classification of Movement analysis data for the early diagnosis of a developing spasticity in newborns with infantile cerebral palsy
Early diagnosis of a developing spasticity in preterm and full term newborns with infantile cerebral palsy is presently based on visual assessment by the attending physician. Diagnosis is based on observations of the spontaneous motor activity of the baby. The reason for the introduction of a new procedure, which is characterized by the subjective impressions of the examiner, is the absence of a methodology for the objective evaluation of spontaneous motion. Thus, a procedure for the objective evaluation of spontaneous motor activity in newborns has been developed at the Helmholtz-Institute.
Methods and Patients
In a first step, 3D motion analysis (Fig.1) was used to record all movements. Information about the factors used for subjective assessment was gathered from experienced neuropediatricians. Those characteristics of the movement patterns which form the basis for visual assessment of a baby’s movement and describe the differences between healthy and affected subjects were identified. Algorithms which reflect the characteristics of motion were developed to extract 53 quantitative parameters from the patient’s 3D movement data. These parameters refer to movement speed, trajectory smoothness, periodicity, range of motion, acceleration and distance between trajectories. Each of these parameters alone does not permit a conclusive statement to be made about the patient being at risk of developing a movement disorder or not. Therefore, to find an optimal combination of parameters, cluster analysis based on Euclidian distances was used. The optimal parameter combination was subsequently used to classify the subjects’ movement into the almost homogeneous classes labelled “healthy” or “at risk” by using an adequate classification procedure. Classification was executed using the quadratic discriminant analysis with 8 parameters. Here, the selected parameter values of a measurement trial are compared with all corresponding parameter sets in a database for which the classification of each parameter set – “healthy” or “affected” - is already known. The classification for the parameter set showing the greatest similarity with the measurement trial is then adapted to the measurement trial itself. The patient and norm collectives consisted respectively of 26 and 66 measurements of preterm and full term newborns.
The classification algorithm allows a reliable differentiation between healthy and affected subjects based on objective data sets from 3D movement analysis. An overall detection rate for previously unclassified subjects of 73% was attained. This value is expected to rise with increasing patient and norm collective database size.
Discussion and Conclusion
Utilising 3D motion analysis and the before mentioned classification, an objective evaluation of spontaneous motor activity in newborns becomes possible. The methodology presented permits the risk prediction of a developing spasticity in newborns.