M. Taylora, T. Kantera, D. Baltab I. G. Porcoc, P. S. Lumd, U. Della Crocec,d, M. M. Schladend
aChildren’s National Hospital, Washington, DC, USA, bPolytechnic University of Turin, Turin, Italy, cUniversity of Sassari, Sassari, Italy, dThe Catholic University of America, Washington, DC, USA
While video movement analysis shows promise for identifying infants at risk for neuromotor impairment [1,2], current gold standard methods require precise setup and time-consuming processing that creates barriers to clinical implementation.[3] There is currently little attention given to paediatric clinicians’ perceived practice improvement needs and how video movement analysis technologies might scale and adapt to day-to-day clinical environments. The purpose of this study was to conduct formative research on clinicians’ perceived utility of video documentation in their practice. Perinatal brachial plexus injury (BPI), where timely diagnosis is of particular concern [4], was our chosen test domain.
We collected video of 14 infant assessments in the biweekly BPI clinic of a major US Children’s hospital. A multidisciplinary team comprised of three clinicians, one each from occupational therapy, physical medicine and rehabilitation, and paediatric neurosurgery, participated in each infant assessment. Video was captured using a Surface Pro 4 tablet with standardized handheld positioning. The multidisciplinary BPI clinic team was debriefed (recorded and transcribed) on the same day as assessments were conducted. Videos were subsequently processed using the MediaPipe machine learning (ML) pipeline to obtain anatomical points of interest (PoI) and to overlay joint angle measures (Figure 1.) on each video frame to serve as an exemplar of possible computational approaches to increase the efficiency of video review. MMS used content analysis to identify key use cases, requirements, and barriers, which were then verified and elaborated by MT, BPI Clinic occupational therapist.
The key driver for use of video in BPI assessment was the affordance it provided clinicians to review and drill down on the subtle quality of an infant’s movement in contrast to the grosser measures documented in the prescribed quantitative scale. The main contexts for use of video were identified as preparing for an infant’s subsequent visits and as an aid to clinical decision making, particularly relative to surgery. Clinicians saw value in automated reporting of relative frequency of BPI pathognomonic, ML-identified movements and postures. Barriers to incorporating video in BPI workflow included time to review, difficulty of video capture given infant unpredictability, the need to setup/troubleshoot equipment, and lack of accommodation of video in the patient electronic medical record.
Video records represent the richest, most accessible repository of movement data in clinical practice - analogous to how MRI provides comprehensive structural information, but without, in the specific context of infants and BPI, requiring sedation or specialized facilities. Video documentation of clinical assessment further enables a full range of possibilities for use, from human review to computational analysis and reporting. Addressing workflow integration barriers represents the critical first step toward unlocking video's potential for computational enhancement in clinical practice. Fostering the routine use of video in clinical practice provides a foundation for building computational tools that authentically address clinicians’ needs and are more likely to improve patient outcomes.
[1] Disselhorst-Klug et al. Experimental Brain Research 2012; 212: 305-313.
[2] Mazzarella et al. Sensors 2020; 20:7312.
[3] Koster et al. Experimental Brain Research 2025; 243(6):153.
[4] O’Shea et al. Plastic Surgery 2025; 22925503241301719.
To be presented at the XXV CONGRESSO NAZIONALE SIAMOC 2025 – Cagliari, 1-4 ottobre 2025