Using Artificial Intelligence to Screen for Hip Dysplasia

Initial screening for Developmental Dysplasia of the Hip (DDH) in Australia is performed most often by general practitioners and paediatricians shortly after birth and by maternal child health care nurses (MCHN) throughout the first year of life. These physical examinations consist of the Ortolani and Barlow tests and the examination of the thigh and gluteal creases. A recent meta-analysis reported the sensitivity of these tests as 36%, which indicates there is potential for a large proportion of cases to go undetected when solely relying on these examinations. Moreover, there currently are no formalised processes by which standards of practice are taught, assessed, or maintained. Thus, there is a clear need for a less operator-dependent screening protocol that can be performed within the current models of infant care. While some countries utilise universal ultrasound screening, this too is limited by access to care, as devices are not portable and thus cannot be used in current care models. Furthermore, it requires a specialist operator, substantially increasing cost. The screening program's limited nature, combined with the need for more consensus among international healthcare providers regarding the best method for managing DDH, has produced highly mixed clinical practices.

One part of the solution is optimising screening protocols for DDH in existing care models. Each state in Australia has established MCHN care protocols that provide access care for young children. While physical screening for DDH in these visits is standard practice, there remains considerable scope for improvement in the accuracy and reliability of these screening methods. Selective screening relies on several clinical associations with DDH to identify which patients receive ultrasound screening. Still, it has been shown to detect only 50% of infants with dysplasia. The MCHN screening program relies on clinical examination alone to detect dysplasia, an inferior identification method. Universal screening has a higher rate of detection of dysplasia but is expensive, single point in time (so misses the development of dysplasia) and results in higher levels of treatment.

A possible solution is portable artificial intelligence (AI)-augmented ultrasound. Recently technology has been developed to support a portable ultrasound device to screen DDH that uses AI-enabled technology to screen for DDH rapidly and accurately. Prior data has demonstrated that physicians and nurses could operate the device following training from expert sonographers. With its low-cost and ease of operation (with simple training) by healthcare providers such as MCHNs, it could significantly augment the physical screening. Thus, there is clear potential for an affordable, repeatable, and accessible screening methodology to be translated into clinical care. Initial Canadian data is promising. Pilot data suggests that DDH detection rates with this technology is on par with the detection rates of orthopaedic specialists. However, as this study was performed in a community setting and only those participants referred to orthopaedic clinics had a standard ultrasound measure performed, this pilot was unable to compare this screening technique with current gold standard diagnostic measures across the whole cohort, nor determine device sensitivity or predictive values. To demonstrate that this technology is fit for purpose, it is imperative that the rate of false negatives is also understood, as this is what will lead to late presentation, - which is what screening ultimately endeavours to prevent. Moreover, in an Australian context an important consideration in a wider roll-out is whether this technology would be accepted for uptake by clinicians and parents.

The proposed project will seek to gather pilot data to assess the validity and feasibility of this technology within a population of infants aged 4-20 weeks flagged at risk for DDH and referred to the Royal Children's Hospital. This will enable the recruitment of a sufficient number of cases of DDH to determine the sensitivity of the device. While the sensitivity and specificity of the device in this at-risk population may not be generalizable to the wider community the information gathered will then inform and refine a larger study of this technology in a community setting such as tertiary (birthing hospitals) and primary (MCHN clinics) care. If it can be demonstrated that it is feasible to implement this technology into existing care models, there is clear scope for this technology to revolutionize DDH screening. Thus, this project seeks to determine how well the device performs (sensitivity, specificity and predictive value) and the the clinical acceptability of this measure within the patient population.

• Factors associated with differences in device sensitivity [Time frame: 1 day, all data will be collected from day of scan]
• Factors associated with differences in device specificity [Time frame: 1 day, all data will be collected from day of scan]
• Factors associated with differences in device positive predictive value [Time frame: 1 day, all data will be collected from day of scan]
• Factors associated with differences in device negative predictive value [Time frame: 1 day, all data will be collected from day of scan]
• Device operator reliability in performing successful scans [Time frame: 12 months or entire study duration]
• Acquisition of successful scans [Time frame: 12 months or entire study duration]
• Time taken to acquire scan [Time frame: 1 day, calculated at time of scan]
• Caregiver perspectives on their infant undergoing the artificial intelligence augmented ultrasound [Time frame: 1 day, caregivers will be asked to complete immediately following the scan]
• Operator perspectives on performing the artificial intelligence augmented ultrasound [Time frame: At the conclusion of their involvement in the study device (up to 12 months)]