Automatic Characterization of the Neonatal Hip Using Ultrasound Imaging

STUDY OBJECTIVES To investigate the use of a novel approach for accurate and automatic extraction of bone, cartilage and muscle surfaces from 3D US images, and to automatically characterize the infant DDH hip. To evaluate the performance of the 3D US-based approach, we will be comparing the results with regular 2D US images of the same patients from the radiology archives at our institution.

HYPOTHESIS We hypothesize that automatically extracted bone/cartilage boundaries can provide useful information about DDH, enable more reliable and reproducible diagnoses, and reduce false diagnostic positives and negatives of DDH. Using 3D US may reveal clinically significant disease or dysplasia metrics that have the potential to change diagnostic procedures.

AIMS:

1. To determine if automatically extracted 3D bone/cartilage surfaces provide useful information (i.e., disease or dysplasia metrics) during the assessment of DDH and provide better visualization of the infant hip anatomy in comparison to regular 2D US-based DDH assessment.
2. To determine if diagnosing DDH using the 3D US images reduces the operator dependent false positive and false negative rates.
3. To determine if 3D US-based dysplasia metrics are as valuable in tracking disease progression as 2D US-based dysplasia metrics.

RESEARCH METHOD Parents and their infants that have been referred to the Orthopaedic Clinic at our institution for suspected or diagnosed DDH will be approached by the Orthopaedic Department's research staff to participate in the study. We are hoping to reach an n=75 as we expect 33% will be lost to follow up. Once eligible infants have been identified by clinical staff, the research staff will outline the study and will obtain informed consent. The parents will be informed of the four additional scan sessions that we would like to collect. Two separate orthopaedic surgeons will each perform one of the 3D and one 2D scan sessions, so as to assess inter-rater reliability. In order to assess intra-rater reliability, each scan session will consist of 2 successive scans.

A major goal of this study is to show that the automated image extraction/analysis by 3D US can reproduce the findings of an expert clinician trained in DDH diagnosis. The 3D images will be processed by using the confidence-weighted local-phase feature based image processing method. Accordingly, the dysplasia metrics can be directly compared between two specialists to determine the feasibility of this approach. Should this method prove to be efficacious, further study could potentially investigate whether this aspect of patient care could be made the responsibility of nurse practitioners.

A secondary goal of this study is to assess how 3D US compares to 2D US in tracking DDH progression. In order to investigate this, we will be giving parents the option to opt in to a second set of US scans if it is necessary for their infant to return to clinic for follow-up treatment. This will not be a mandatory criterion for inclusion into the study. However, as current clinical practice involves checking for DDH progression with 2D US scans, it will provide additional information that could be valuable for comparing the two scanning methods in tracking DDH progression. This option will be presented to parents in the form of a check box on the consent form for them to opt in if they so wish.

DATA ANALYSIS The reliability and accuracy of the 3D image-derived dysplasia metrics will be compared with the standard 2D B-mode US images in terms of variability within inter-rater and intra-rater measurements. For statistical significance of improvement in the diagnostic procedure, we require at least 50 infants with at least one affected hip or suspected affected hip with DDH. We have asked for a recruitment size of 75 as we believe that at least 33% will be lost to follow up. This will be achieved by performing a comparison study between the 2D and 3D US images collected for each patient at each clinic visit. Orthopaedic surgeons from our institution will be asked to identify important features from both (processed and un-processed) US images. For 2D images, the surgeons will measure α and ß angles and percent coverage of the femur. Intra- and inter-rater reliability will be calculated using kappa scores. For 3D images, profiles of α and ß angles and 3-dimensional percentage coverage of the femur (i.e. volume coverage instead of surface coverage), will be measured. These manual measurements will be compared to the measurements determined by automatic image extraction and analysis to test for improvement in diagnostic procedure.