Original Contribution3-D Ultrasound Imaging Reliability of Measuring Dysplasia Metrics in Infants
Introduction
Developmental dysplasia of the hip (DDH) is the most common hip disorder in infants, and refers to a spectrum of hip abnormalities ranging from mild dysplasia with a stable hip through subluxation to total hip dislocation (Sewell et al. 2009; Dornacher et al. 2010; Gulati et al. 2013; Shorter et al. 2013; Jaremko et al. 2014; Pollet et al. 2017). Current methods for measuring the degree of dysplasia in newborns using ultrasound (US) imaging are notoriously unreliable (Jaremko et al. 2014; Orak et al. 2015). Our recent systematic review (Quader et al. 2018) identified that the most commonly used dysplasia metric, α2D (defined as the angle between the vertical cortex of the ilium I and the acetabular roof A; Fig. 1b) has considerable inter-exam variability—7° standard deviation—which is problematic because its range between healthy and dysplastic hips is only 17°.
This high variability can lead to non-trivial under- and over-treatment rates—29% (Imrie et al. 2009) and 38% (Shipman et al. 2006), respectively. Under-treatment a failure to diagnose and treat DDH in infancy can result in costly corrective surgical procedures later in life (e.g., open reduction of dislocated hips with or without pelvic or femoral osteotomy in infants; Shipman et al. 2006) and total hip replacement procedures in adults (Price and Ramo 2012). On the other hand, over-treatment of DDH can lead to adverse consequences such as avascular necrosis (the death of bone tissues owing to a lack of blood supply), psychological stress on infants and parents and dislocation owing to repeated dynamic hip assessments (Patel and Canadian Task Force on Preventive Health Care 2001; Shipman et al. 2006).
In an effort to improve reliability and standardize US-based DDH assessments, Germany has established special commissions which have established a checklist for hip sonographers to follow (Graf et al. 2013). In 2011, one German commission revoked the licenses of up to 43.7% of the hip sonographers in eight German states owing to deficiencies in their protocols (Tschauner and Matthiessen 2012). Later, in two separate studies, Jaremko et al. (2014) and Kolb et al. (2017) discovered that the poor reliability of US-based DDH diagnosis is primarily owing to attempting to use a 2-D US image for diagnosing a 3-D hip morphology. A 2-D US-based dysplasia metric is sensitive to the orientation and placement of the 2-D US transducer during image acquisition, which gives rise to variability in diagnosis (Jaremko et al. 2014; Kolb et al. 2017) (Fig. 1d–f). We hypothesize that by using 3-D volumes that capture the entire femoral head and its neighboring structures in a normal infant hip, we can consistently extract the acetabular and femoral head morphology regardless of location/orientation during acquisition. Dysplasia metrics derived from 3-D hip morphology should therefore be significantly less variable than the conventional dysplasia metrics derived from 2-D US images.
Recently, two simultaneous studies proposed using 3-D US to reliably diagnose DDH; our group (Quader et al. 2016) proposed the α3D metric (defined as the angle between the planar approximations of the vertical cortex of the ilium I and the acetabular roof A; Fig. 1g), and Hareendranathan et al. (2016) proposed the acetabular contact angle (ACA, the angular separation between the surfaces of A and I).
The method of Hareendranathan et al. (2016) involves a slice-by-slice graph search-based process that requires manually selecting three seed points in each of the 2-D US slices in a 3-D US volume and manually separating A from I, which requires valuable clinician time and introduces within-image measurement variability of around 1° and inter-exam variability of around 4° (Mabee et al. 2016). A recent multi-center study (Zonoobi et al. 2018) showed that the semi-automatic ACA computing method of Hareendranathan et al. leads to slightly more accurate DDH diagnosis than 2-D US-based approaches; however, the improvement was not statistically significant. This lack of statistically significant improvement in diagnosis reliability is also consistent with our recent systematic review, finding that no statistically significant reduction in variability of repeated ACA measurements compared with that of α2D (Quader et al. 2018).
In contrast, our original α3D metric (Quader et al. 2016) was completely automatic, and demonstrated reduced inter-exam variability (standard deviation of 2.19°). Given a 3-D US image, our method would approximate the femoral head as a spherical object using a sphere fit based on M-estimator Sample Consensus (MSAC) to localize the femoral head, use that approximated femoral head as a prior for localizing I and A, use a Radon transform-based plane fit to approximate the planes of A and I and then calculate α3D as the angle between the planes of A and I. We denote the α3D extracted using this method as α3 D,MR.
While we found a modest reduction in intra-exam variability with α3D,MR using our previous method (Quader et al. 2016) compared with α2D (Graf 2006—around 29% reduction, from 3.08° to 2.19°—there were some additional limitations. The original α3D,MR extraction extracting method assumed that I boundaries would be present superior to the center of the femoral head, and that the width of I along the sagittal axis would be half of the diameter of the femoral head. It also assumed that the femoral head is spherical. Because femoral heads can be aspherical in shape, with a tendency for decreased sphericity in hip dysplasia (Rosenberg et al. 2017), presuming that the femoral head is spherical and that I can be localized from an approximate location of the femoral head are inaccurate priors. Relying on such ad hoc priors makes our previous method vulnerable to variation in hip morphology across the patient population. In this article we provide important method modifications to our existing work (Quader et al. 2016) by primarily removing inaccurate priors (i.e., sphericity of the femoral head and location of A and I with respect to the femoral head).
In addition to the α metric, the American College of Radiology (2012) recommends using the femoral head coverage (FHC: the ratio of the femoral head portion medial to I to the entire femoral head Fig. 1c). We have previously presented an automatic method to estimate FHC3D using random forest-based femoral head localization and an MSAC-based ilium fit (Quader et al. 2017a). We denote the FHC3D extracted using this method as FHC3D,RM. Similar to α3D,MR, automatically estimating FHC3D,RM involved using ad hoc priors to localize I. We remove the use of such priors in this article. Furthermore, we investigate the performance of our algorithms on an extended data set of real infant hip examinations.
Our research questions in this paper are as follows:
- •
Will allowing for a more realistic assessment of the femoral head shape and less stringent presumptions about the locations of A and I with respect to the femoral head result in more reliable and robust dysplasia-metric estimation?
- •
Will α3D and FHC3D estimated with our current method be significantly more reliable than when estimated with our previous methods (Quader et al. 2016, 2017a)?
- •
Will α3D and FHC3D estimated with our current method be significantly more reliable than conventional α2D and FHC2D metrics evaluated from 2-D US images?
A list of all acronyms used in this paper is provided in Table 1.
Section snippets
Materials and Methods
We first present the protocol we used in collecting US data from infant hips, and then we present our approach to automatically estimating two key dysplasia metrics, α3D and FHC3D, from 3-D US images. Finally, we present our validation scheme to evaluate the reliability of the extracted α3D and FHC3D in the clinical data set we collected.
Variability of dysplasia metrics
Figure 6 shows an example of extracted dysplasia metrics from two repeated US scans of a hip (Fig. 6a, 6b), extracted α3D,MR and FHC3D,RM metrics from two repeated 3-D US scans of the same hip (Fig. 6d, 6e) and extracted α3D,RSM and FHC3D,RSM metrics from the same two 3-D US scans (Fig. 6g, 6h). Manually aligning the two different scans revealed that the two 2-D images (Fig. 6c) were considerably different, that the two bone boundaries extracted using the MR method (Fig. 6f) had less mismatch
Variability of metrics
The variabilities in the automatically extracted 3-D US-based dysplasia metrics using our RSM method were significantly lower than those of their 2-D US-based counterparts (∼75% reduction for α and ∼ 65% reduction for FHC). These reductions in variability suggest that probe position variation has a much larger effect on variability in the dysplasia metrics than manual processing of the 2-D US (∼10% improvement with automatic image processing in 2-D US; Quader et al. 2017b).
We found that the
Conclusions
In this article, we present novel 3-D hip morphology-based dysplasia metrics, along with automatic methods for extracting those dysplasia metrics from 3-D US B-mode images. We show that these dysplasia metrics are significantly less variable than their 2-D counterparts (∼65% lower variability for FHC3D,RSM and around 75% reduced variability for α3D,RSM). This suggests that the 3-D morphology-derived dysplasia metrics could be valuable in improving reliability in diagnosing DDH, which may lead
Acknowledgments
We thank the Canadian Institutes of Health Research (grant CPG-140180), the Natural Sciences and Engineering Research Council of Canada (grant CHRP 478466-15), the Institute for Computing, Information and Cognitive Systems and the Centre for Hip Health and Mobility.
Conflict of interest
We certify that they have NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers' bureaus; membership, employment,
References (39)
- et al.
Ultrasound confidence maps using random walks
Med Image Anal
(2012) - et al.
Ultrasound bone segmentation: A scoping review of techniques and validation practices
Ultrasound Med Biol
(2020) - et al.
Relative risk and incidence for developmental dysplasia of the hip
J Pediatr
(2017) - et al.
Prevention of hip dysplasia in children and adults
Orthop Clin North Am
(2012) - et al.
Automatic evaluation of scan adequacy and dysplasia metrics in 2-D ultrasound images of the neonatal hip
Ultrasound Med Biol
(2017) - et al.
MLESAC: A new robust estimator with application to estimating image geometry
Comput Vis Image Underst
(2000) AIUM Practice Guideline for the Performance of an Ultrasound Examination for Detection and Assessment of Developmental Dysplasia of the Hip
J Ultrasound Med
(2013)- et al.
Real-time extraction of local phase features from volumetric medical image data
Random forests
Mach Learn
(2001)- et al.
Histograms of oriented gradients for human detection
Bone enhancement filtering: Application to sinus bone segmentation and simulation of pituitary surgery
Comput Aided Surg
Early radiological outcome of ultrasound monitoring in infants with developmental dysplasia of the hips
J Pediatr Orthop B
Multiscale vessel enhancement filtering
Hip sonography: 20 years experience and results
Hip Int
Hip sonography update: Quality-management, catastrophes—tips and tricks
Med Ultrason
Developmental dysplasia of the hip in the newborn: A systematic review
World J Orthop
Automatic extraction of bone surfaces from 3D ultrasound images in orthopaedic trauma cases
Int J Comput Assist Radiol Surg
A technique for semiautomatic segmentation of echogenic structures in 3D ultrasound, applied to infant hip dysplasia
Int J Comput Assist Radiol Surg
Toward automatic diagnosis of hip dysplasia from 2D ultrasound
Cited by (9)
Ultrasound-based 3D bone modelling in computer assisted orthopedic surgery–a review and future challenges
2024, Computer Assisted SurgeryPaediatric hip ultrasound THE INTERPRETATION OF QUALITY, GEOMETRY, AND MORPHOLOGY
2023, Bone and Joint JournalApplications and problems of artificial intelligence technology for ultrasonic screening of developmental dysplasia of the hip
2023, Chinese Journal of Pediatric SurgeryFemur reconstruction in 3D ultrasound for orthopedic surgery planning
2023, International Journal of Computer Assisted Radiology and SurgerySegmentation of measurable images from standard plane of Graf hip ultrasonograms based on Mask Region-Based Convolutional Neural Network
2023, Joint Diseases and Related Surgery