Improving Neonatal Hip Screening With Artificial Intelligence

A recent study by our team showed that at The Royal Women's Hospital, Melbourne, screening at birth using the standard combination of neonatal hip examination and risk-based referral for ultrasound failed to detect 52% (n=100) of cases of Developmental Dysplasia of the Hip (DDH) and that 98.5% (n=2,637) of infants undergoing a screening ultrasound, due to perceived increased risk, do not have DDH. Further, this research team replicated in a regional setting at University Hospital, Geelong in Victoria (n=1,207), that 55.6% of cases of DDH were missed, and of those sent for diagnostic ultrasound scans 92.5% did not have DDH (unpublished data). Together, this means many infants are being scanned, but an unacceptably high proportion of cases are still being missed.

Late detected dysplasia is often resistant to conservative treatment. This form of dysplasia is unpredictable in its presentation and may require surgical intervention to obtain a contained and stable joint. Such patients are at a higher risk of developing degenerative hip disease in early adult life and can suffer considerable disability, often failing to reach their full potential. Thus, many initiatives have been taken to improve our current screening programs, including clinical education programs, streamlined access, and incorporation of hip examinations into child health assessments. However, none of these initiatives has effectively reduced the rate of late detection of dysplasia. Despite selective screening protocols being in place, the incidence of late-diagnosed DDH has increased in South Australia, showing a significant rise from 0.22 per 1000 live births (1988-2003) to 0.77 per 1000 live births (2003-2009).

One part of the solution is optimising screening protocols for DDH in existing care models. A possible solution is utilising artificial intelligence to aid in screening decisions. One such new tool is the Exo Iris, a portable ultrasound device supported by real-time AI-augmented analysis to screen for hip dysplasia. Importantly, use of this device does not require extensive training and could be performed by midwives or paediatricians in standard neonate hip examinations. Initial work has shown that AI could successfully identify the standard plane, make measurements, and classify the hip as normal or abnormal. Scans are simple to conduct, add little time to the overall consultation and are non-invasive without the use of ionising radiation. Importantly, non-experts can easily be trained to use the technology; they are cost-effective and can be used in any clinical environment connected to a standard tablet.

Initial Canadian data suggests that DDH detection rates suggests that artificial intelligence (AI) analysis for hip dysplasia are on par with orthopaedic specialists. Of the infants flagged for follow-up there were 6 subsequently referred to specialist clinics after repeat scan and all were treated for DDH (5 harnessed, 1 surgical intervention). Of these the six infants detected, only two presented with well documented risk factors for increased risk of DDH (female sex, Indigenous, breech, family history), which may not have been detected without initial portable ultrasound screening.

Further to this, Retuve is a new open-source software tool that uses AI-analysis to measures standard indices on hip ultrasound images collected from any manufacturer's ultrasound probe, which can help users make hip screening decisions. This platform generates novel imaging parameters beyond current standards that may also be helpful in further understanding undetected late presentations. However, as this is a relatively new tool there has been little research to fully evaluate its performance and its potential utility as a screening tool.

There is clear scope for this technology to revolutionise screening in Australia by reducing the number of cases of DDH missed and the number of costly conventional ultrasound scans; however, there is limited data assessing the feasibility and accuracy of the new AI analysis strategies for DDH screening in the context of neonatal screening. To date, although the Exo Iris and embedded hip AI software is indicated for use in infants 0-6 months there is a dearth of data examining neonates, and no data exists evaluating it in the context of selective neonatal screening.

This study will provide foundational pilot evidence on the accuracy and feasibility of using the Exo Iris probe and an alternate AI software platform as part of standard neonatal screening. In addition to this, long-term follow-up as part of the VicHip parent study will aid in understanding potential parameters associated with missed presentations. To do this, 100 infants will be recruited from Monash Medical Centre, Victoria, Australia and will undergo both the AI-ultrasound scan that will be analysed both by the embedded software and Retuve and results from the AI analyses will be compared to a 6-week full diagnostic ultrasound.