Assessment of General Movements

General Movements are spontaneous movement patterns in infants aged 3 to 5 months. Abnormal, missing or sporadic movements are associated with a high risk of later neurological dysfunction (especially cerebral palsy). At the same time, the occurrence of normal movement patterns is highly likely to indicate an inconspicuous neurological development in children. For these reasons, the detection and classification of general movements in infants is extremely important so that appropriate intervention can be initiated at an early stage.

In close cooperation with Prof. Ute Thyen (Head of the Neuropaediatrics and Social Pediatrics Center, University Medical Center Schleswig-Holstein in Lübeck), we are developing a learning-based pattern recognition platform that will automate the General Movements examination using algorithms. As data input, we use images from a normal (RGB) video camera, a depth camera (Microsoft Kinect or suitable smartphone) and sensors (e.g. an accelerometer) built into clothing. Following the learning-based approach of pattern recognition, the classification algorithms of the platform are trained with sensor data, which are manually evaluated by the medical experts of our consortium regarding general movements. The results of the automated recognition and classification of movement patterns will be presented to the treating physicians on site using an app and, if necessary, transferred telemedically to a specialized medical facility. Another important goal for us is also to significantly improve the availability of this examination so that it can be made available nationwide (even in rural areas).

Methodologically, we are working on three pattern recognition approaches in this context. First, we investigate the use of classification methods based on manually selected feature vectors. Second, we develop a hierarchical recognition algorithm that first performs the classification separately along the assessment criteria used by medical experts (complexity, variation, fluency) and only finally fuses the results. Third, we implement deep learning algorithms for the recognition of general movements.

Contact person

Prof. Dr.-Ing. Marcin Grzegorzek