New technology looks set to help families access timely screening and early intervention.
A three-minute video of your baby may soon be all that you need to screen for cerebral palsy, according to the Murdoch Children’s Research Institute (MCRI).
In a recent study, videos of 327 infants aged 12-18 weeks of age were assessed by a “pose-estimation” algorithm, embedded in a smart-phone app.
The video analysis program was found to have a balanced accuracy of 70% in identifying the abnormal or absent fidgety general movements suggestive of cerebral palsy.
The Baby Moves smartphone app is currently available to parents and health professionals to upload videos of infants to be reviewed by a trained assessor. However, being able to automate this assessment using the algorithm may help bridge the gap that exists now in accessing timely treatment for cerebral palsy children.
The algorithm automates the currently used, specialist-led General Movements Assessment for identifying high-risk children and uses AI to analyse videos submitted, said Dr Elyse Passmore, clinician scientist fellow in the Developmental Imaging Group at the Murdoch Children’s Research Institute.
“In validation studies, our AI system tracked 18 key body points and achieved a balanced accuracy of 70% in identifying infants with abnormal or absent fidgety general movements,” she said.
“Although sensitivity (76%) and specificity (65%) are lower than those of trained assessors (sensitivity 97%, specificity 89%), our tool significantly enhances accessibility and scalability.”
According to the study, there was no difference in accuracy due to camera resolution. Currently, children suspected of having a movement disorder are screened by undergoing a General Movements Assessment (GMA) which is conducted by a specialist assessor in person.
The General Movements Assessment can be utilised from 12 weeks of age, and it is known that targeted intervention for abnormal motor development is crucial in the first six months of life for both the child’s cognitive development and parental wellbeing.
However, the study authors noted that this assessment was not routinely offered in hospitals or primary care and could only be conducted by assessors with specialised training, so access was often significantly delayed.
In Australia, the average age of cerebral palsy diagnosis is 19 months, with only 21% of infants with cerebral palsy being diagnosed before six months of age. The authors noted that “many infants miss a crucial window for early intervention”.
“We emphasise that both the [General Movement Assessment] undertaken by a trained assessor and our automated algorithm are assessment tools and are not diagnostic instruments,” said Dr Passmore.
“With false positives at 33%, further refinement and validation efforts of our algorithm are now underway.
“We are also collaborating with Generation Victoria (GenV), Australia’s largest longitudinal birth cohort. This large-scale population study will help us to analyse over 4000 additional infant movement videos and help fortify the robustness of our algorithms.”
