Tracking General Movements

General movements (GMs) are spontaneous movements of the entire body observable from 9 weeks of gestation until at least five months postterm age. The assessment of GMs, the GMA, is based on the visual evaluation of normal vs. abnormal movements and has proven to be a reliable tool to predict neurodevelopmental outcomes. Trained observers show high agreements in rating GMs, but their accuracy declines when the GMA is applied infrequently. Certain changes in the quality of GMs have been linked to neurodevelopmental disorders. However, the pattern of poor-repertoire GMs (i.e. a monotonous movement pattern) is still of low diagnostic power and warrants detailed study. To date, only little is known about an observers visual and perceptual processes during GMA. Using eye tracking to observe the observer, this project will help to gain new insights into the clinical reasoning processes involved during GMA. With our innovative approach, we aim to: (1) learn more about the not yet fully understood poor-repertoire GM pattern [enhance diagnostics]; (2) further develop GMA training methodologies [enhance clinical training]; (3) implement a globally accessible database for clinical GMA training [sustainability]. Knowledge gained in this project will in addition support the development of computer-based analyses of motor patterns. As the GMA is easy to apply, non-invasive and cost-efficient, it has been recommended for broader application, also in low- and middle-income countries and remote areas, where access to modern imaging techniques is limited. Observing the observer will (a) contribute to improve clinical diagnostics of the GMA by redefining the role of poor-repertoire GMs; (b) advance training of GMA by establishing an innovative training method and a database for clinical GMA training; and (c) set the basis for future endeavours to improve pattern recognition algorithms for automated GMA.

Our project contributed to the better understanding of spontaneous motor functions, mainly general movements, from fetal life to the early postnatal period. These movements are continuously present without sensory input and can be assessed with the General Movement Assessment (GMA). GMA is a widely used method to detect early neurological impairments that relies on highly trained specialists. Our project substantially contributed to the growing field of developing automated measurement and classification tools of infant motor functions and combined medical expertise with computer vision, eye-tracking, and data visualisation technologies. We developed tools that automatically track and classify infants' movements in video recordings, creating a virtual skeleton and quantifying metrics from sensor fusion approaches (video, pressure mat, gyroscopes) such as movement space, speed and angles. We analysed different types of spontaneous movements, including writhing movements, which occur shortly after birth, and fidgety movements, which appear later in early infancy building models for movement classification. Changes in these movements are key indicators of altered neuromotor development, and with our tools we added to the current research on objectively measuring and visualizing early adversities. With this project we provided thus a contribution to the future role of computer vision and machine learning in GMA. At the same time, we built an innovative eye-tracking system capable of dynamically following body regions of interest and linking them directly to the observer's gaze. Unlike previous methods, this approach allows automated, detailed analysis of how observers look at infants' movements in real time. Experts showed distinct gaze behaviours when evaluating physiological and pathological movements, providing the first quantitative evidence of how medical expertise shapes visual perception in early neurological assessment. This marks an important starting point for developing new GMA support- and training-tools. In the long run these findings have the potential to improve GMA training worldwide, including in countries with currently limited access to expert trainers Beyond analysis, we created interactive dashboards for expert discussion and debriefing. Building on these insights, we developed a data-driven training concept for novices, combining expert movement annotations, replay of expert gaze patterns, and real-time feedback. This concept lays the foundation for more effective, evidence-based training in GMA. Moreover, we established a replicable, standardised methodology for preparing GMA stimulus materials for eye-tracking research, enabling (international) multi-site data collection with consistent stimulus presentation. The project also produced mobile and AI-supported solutions for early detection, including publications on General Movement apps. By transforming expert knowledge into measurable patterns, our work lays the foundation for more objective, scalable, and digitally supported early diagnostics, potentially improving clinical outcomes, expanding access to specialised expertise worldwide, and shaping the future of infant healthcare.