Video-based Body Parsing for Neonatal Body-parts Actigraphy: A Clinical Study in NICU

Area of Science:

• Biomedical Engineering
• Developmental Neuroscience
• Clinical Neonatology

Background:

• Video-based monitoring offers a flexible approach to assessing infant activity.
• Actigraphy serves as a key indicator for evaluating neonatal nervous system function.
• Reduced movement complexity and variability in infants can signal central nervous system damage, such as cerebral palsy.

Purpose of the Study:

• To develop and validate a new method for extracting body-parts actigraphy from video data.
• To assess the clinical significance of detailed body-part movement analysis in newborns.
• To improve neonatal monitoring for developmental status and neurological conditions.

Main Methods:

• Utilized a state-of-the-art body parsing model, retrained on the YouTube dataset, for accurate infant body segmentation.
• Employed PixFlow and Farneback optical flow algorithms to calculate body-parts actigraphy.
• Validated the method on a clinical dataset of newborn infants.

Main Results:

• The body parsing model demonstrated accurate segmentation of infants within the clinical dataset.
• High Pearson correlation coefficients (0.93-0.98) were achieved for motion signals between PixFlow and Farneback algorithms across different body parts (head, arms, torso, legs).
• The proposed method effectively quantifies infant movement complexity and variability.

Conclusions:

• Body-parts actigraphy provides a more detailed assessment of infant movement compared to traditional methods.
• This technique holds significant potential for enhancing neonatal monitoring applications, including sleep analysis and prognosis assessment.
• The findings support the use of video-based body-parts actigraphy for early detection of neurological impairments and monitoring developmental progress.