Automatic bone segmentation and bone-cartilage interface extraction for the shoulder joint from magnetic resonance images

Abstract

We present a statistical shape model approach for automated segmentation of the proximal humerus and scapula with subsequent bone-cartilage interface (BCI) extraction from 3D magnetic resonance (MR) images of the shoulder region. Manual and automated bone segmentations from shoulder MR examinations from 25 healthy subjects acquired using steady-state free precession sequences were compared with the Dice similarity coefficient (DSC). The mean DSC scores between the manual and automated segmentations of the humerus and scapula bone volumes surrounding the BCI region were 0.926 ± 0.050 and 0.837 ± 0.059, respectively. The mean DSC values obtained for BCI extraction were 0.806 ± 0.133 for the humerus and 0.795 ± 0.117 for the scapula. The current model-based approach successfully provided automated bone segmentation and BCI extraction from MR images of the shoulder. In future work, this framework appears to provide a promising avenue for automated segmentation and quantitative analysis of cartilage in the glenohumeral joint.

Automatic segmentation of the glenohumeral cartilages from magnetic resonance images

“…The bone segmentation with the FSM was highly successful. Compared to previous results with standard SSMs, 44 there was an improvement in accuracy for the humerus (mean DSC from 0.926 to 0.978, MASD from 0.943 to 0.255 mm) and scapula (mean DSC from 0.837 to 0.925, MASD from 1.027 to 0.533 mm). For comparison, a recent study presenting segmentation results of normal humeral heads from MR images achieved a mean accuracy of 0.9217 in 2D.…”

“…54 Similar to the bone segmentation, there was an improved BCI extraction for the humeral and glenoid cartilages with mean DSC scores of 0.835 and 0.856, respectively, compared to 0.806 and 0.795 with standard SSMs. 44 An initial probability of 0.9 was used for BCI extraction, resulting in intentional over-segmentation of the BCI. This was done to avoid undersegmentation in BCI, which may lead to unrecoverable cartilage in downstream processing.…”

Identification of locations susceptible to osteoarthritis in patients with anterior cruciate ligament reconstruction: Combining knee joint computational modelling with follow-up T1ρ and T2 imaging

“…Generation of a subject-specific computational model requires a lot of manual work and time in segmentation of soft tissues, meshing and making models to converge. In future studies, the methodology presented here should be coupled with semi-automatic or fully automatic segmentation techniques (Chandra et al, 2016;Dodin et al, 2010;Folkesson et al, 2007;Lee et al, 2014;Liukkonen et al, 2017b;Paproki et al, 2014;Shan et al, 2014;Tamez-Pena et al, 2012;Yang et al, 2015;Yu et al, 2016) and with automated meshing tools (Rodriguez-Vila et al, 2017). As motion capture systems are not readily available in clinical settings, a simple and fast method should be developed to obtain and implement patient's gait.…”

Segmentation of joint and musculoskeletal tissue in the study of arthritis

“…In comparing different musculoskeletal anatomy, it is important to comment on the extensive use of SSM and atlas-based methods in spine segmentation, particularly for the analysis of lumbar and thoracic intervertebral discs (IVDs) and vertebral bodies (VBs) [77], and 3D segmentations of healthy and herniated intervertebral discs [78]. SSM was also recently used to identify the bone cartilage interface in the shoulder joint [79]. A combined use of SSM and atlas-based approaches has been presented in the segmentation of the quadratus lumborum muscle [80].…”