Engineering and Technology
Da Yeong Kim, MD
Resident
Department of Rehabilitation Medicine, Asan Medical Center
Seoul, Cholla-bukto, Republic of Korea
Eun Jae Ko, MD
Professor
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Joon Hee lee, MD
Resident
asan medical center, university of ulsan college of medicine
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Jong Yoon Chang, MD
Resident
Department of Rehabilitation Medicine, Asan Medical Center
Seoul, Cholla-namdo, Republic of Korea
Sunyoung Joo, MD
Fellow
Incheon St. Mary's Hospital
Incheon, Inch'on-jikhalsi, Republic of Korea
Tae Won Kim, MS
Researcher
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Sumin Kim, BS
Researcher
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Kyung Yong Choi, BS
Physical therapist
Department of Pediatric Rehabilitation Unit, Asan Medical Center
Songpa-gu, Seoul-t'ukpyolsi, Republic of Korea
In Jin Yoon, BS
Occupational Therapist
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Sae Mi Hong, PhD
Speech-Language Pathologist
Department of Pediatric Rehabilitation Unit, Asan Medical Center
Songpa-gu, Seoul-t'ukpyolsi, Republic of Korea
Seung Hak Lee, MD. PhD.
Professor
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Seungwoo Cha, MD
Professor
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
June-Goo Lee, PhD
Professor
Asan Medical Center
Seoul, Seoul-t'ukpyolsi, Republic of Korea
Fine motor development plays a crucial role in both motor and cognitive domains, and early fine motor difficulties are recognized as early indicators of global developmental delay. This study aimed to develop an AI-based video analysis system to quantitatively evaluate the “Grasp of Cereal” item from the QUEST, enabling objective assessment of fine motor function in children with cerebral palsy and developmental delay.
Design:
This prospective cohort pilot study included children with cerebral palsy (CP) and those with suspected developmental delay (DD). All participants performed the “Grasp of Cereal” item from the QUEST, and performances were video-recorded. Hand movements were detected using MediaPipe, refined through SAM2 segmentation, and synchronized with cereal-object detection by YOLOv9 to identify precise hand–object interactions. Agreement between AI predictions and expert ratings was assessed using accuracy, severity-weighted accuracy (SWA), and quadratic weighted kappa (QWK). Clinical variables, including MACS and BSID-II results, were collected.
Results:
The AI model achieved an overall accuracy of 0.73, a SWA of 0.859, and a QWK of 0.446 across all trials. Performance was better in children with cerebral palsy (accuracy = 0.84, SWA = 0.922, QWK = 0.579) than in those with suspected developmental delay (accuracy = 0.38, SWA = 0.659, QWK = 0.200). “Fine pincer” and “pincer” patterns were most reliably identified, whereas “inferior scissor” was often misclassified. Inter-rater reliability between expert therapists was high (accuracy = 0.756, SWA = 0.937, QWK = 0.883), supporting the validity of the ground-truth labeling and demonstrating strong human-level agreement for fine motor posture evaluation.
Conclusion:
This pilot study demonstrated that a deep learning–based AI model can accurately evaluate the “Grasp of Cereal” item from the QUEST, suggesting its feasibility as an objective assessment tool.
