An Automatic Segmentation Method for Regional Analysis of Femoral Neck Images Acquired by pQCT

Abstract

We developed an automatic method for regional analysis of femoral neck images acquired by peripheral quantitative computed tomography (pQCT), based on automatic spatial re-alignment and segmentation; the segmentation method, based on a morphological approach, explicitly accounts for the presence of three different bone compartments: cortical region, trabecular region, and transition zone between cortical and trabecular compartments. The proposed method was applied on 13 femoral neck sections derived from female donors who were undergoing hip replacement surgery for primary degenerative arthritis or fracture, and a typical densitometric and structural analysis was performed both globally and regionally. The proposed segmentation method was quantitatively evaluated by comparing automatic contour and the corresponding manual contours delineated by three operators using metrics based on surface distance (average symmetric distance, ASD) and volumetric overlapping (dice similarity coefficient, DSC). The same approach was used to validate the automatic spatial orientation, considering as metric the difference between manual and automatic angle orientation. Results confirm a satisfactory agreement between automatic and manual performances (ASD < 0.41 mm, DSC > 0.91, orientation difference = 3.61°) and show that globally our algorithm performs very well. Concerning regional analysis application, from our results we can observe that significant differences are present among the four bone quadrants.

Comparative high-resolution pQCT analysis of femoral neck indicates different bone mass distribution in osteoporosis and osteoarthritis

“…Each pQCT scan was analyzed using a novel segmentation algorithm developed in-house to automatically extract bone structure descriptors, in both cortical and trabecular compartments [15]. The segmentation method explicitly accounted for the presence of three areas of interest: the cortical region corresponding to the area between the outer bone contour and the inner cortical bone contour, the trabecular region, within the trabecular contour, and the transition zone between the cortical and trabecular compartments, often located in proximity to the inferior part of the femoral neck where the advanced porosity process of the cortex assumes, at least in part, trabecular features.…”

“…The transition zone is automatically excluded from analysis after segmentation without relying on model assumptions [18]. Quality data of the segmentation process were reported [15]. The new in-house software based on automatic spatial realignment and segmentation provides an additional advantage that might be critical in understanding the OA-OP relationship.…”

“…Ct.P.Th was derived by applying the same algorithm for the estimation of Tb.Th to the porosities (holes) in the cortical region. The Tb.N and Tb.Sp were derived using the following equations, which are typically used in standard histomorphometry [15]: Moreover, in each group, possible age-adjusted correlations between cortical and trabecular descriptors in the whole sample were estimated using multiple linear regression models in which each trabecular parameter was considered as the response, and the corresponding cortical parameter and age were included as regressors. For each test, significance was set at p=0.05.…”

“…Actual scanners display sufficient resolution (70-84 μm) to resolve both the cortical and trabecular architecture of human bone in the three spatial dimensions [15] and provide direct measurements of bone size, which is an independent predictor of bone strength [16], as well as the non-metric descriptors of bone structure that constitute the 3D data background for finite element prediction of bone strength [17]. The pQCT large scanner gantry opening (9 cm) can include the complete femoral neck section for direct analysis of its structure in the same scanning without relying on model assumptions based on the 3D reconstruction of cancellous bone cores [18].…”

Bidirectional local distance measure for comparing segmentations

“…20,24,25 Rizzo et al utilized the minimum distance measure to calculate the average symmetric distance, which is one type of Hausdorff distance. 26 However, the Hausdorff distance is a global distance measure as mentioned above. It is not able to locate the regional discrepancy.…”