Statistical Interspace Models (SIMs): Application to Robust 3D Spine Segmentation

Abstract

Statistical shape models (SSM) are used to introduce shape priors in the segmentation of medical images. However, such models require large training datasets in the case of multi-object structures, since it is required to obtain not only the individual shape variations but also the relative position and orientation among objects. A solution to overcome this limitation is to model each individual shape independently. However, this approach does not take into account the relative position, orientations and shapes among the parts of an articulated object, which may result in unrealistic geometries, such as with object overlaps. In this article, we propose a new Statistical Model, the Statistical Interspace Model (SIM), which provides information about the interaction of all the individual structures by modeling the interspace between them. The SIM is described using relative position vectors between pair of points that belong to different objects that are facing each other. These vectors are divided into their magnitude and direction, each of these groups modeled as independent manifolds. The SIM was included in a segmentation framework that contains an SSM per individual object. This framework was tested using three distinct types of datasets of CT images of the spine. Results show that the SIM completely eliminated the inter-process overlap while improving the segmentation accuracy.

Automatic segmentation of vertebral contours from CT images using fuzzy corners

“…In order to validate and assess the performance, two most commonly used metrics, Dice similarity coefficient (DC) and Hausdorff distance (HD) are used. The quantitative analysis metric is widely used for testing the accuracy of segmentation [37,38]. The dice similarity coefficient is computed as,…”

Automatic segmentation of the spine by means of a probabilistic atlas with a special focus on ribs suppression

“…The results reported by Korez et al and Castro‐Mateos et al were more accurate in the thoracic region but, to evaluate this region, they only included healthy patients. In addition, the development of completely automatic algorithms is still an open problem, being necessary in many cases some degree of user interaction . This has been accomplished with the current method, which is fully automatic.…”

“…The development of methods using prior knowledge of the shape to be segmented is an active field of research. 9 Therefore, most of the developed methods are based on deformable models, [10][11][12][13][14][15][16][17][18][19][20] which make use of this information. However, the availability of this kind of data is not always possible, therefore, other methods that do not require any form of prior knowledge are used, too.…”

“…[24][25][26][27] Level-set methods are particularly appropriate for dealing with different features such as cavities or convoluted areas. However, many of these works either do not segment all thoracic and lumbar vertebrae, [10][11][12][13][14][15]20,[23][24][25][26] or they are not completely automatic, [10][11][12]19,20,23,25 or have been only tested in healthy cases. 18,25 The presence of the ribs is one of the problems for which different methods show less accuracy in the segmentation of the thoracic region.…”

“…This is the reason why the majority of studies have only been focused on this region. [10][11][12]15,20,[23][24][25] Despite that, several studies found in the literature have made a great effort to segment all thoracic and lumbar vertebrae, [16][17][18][19]21,22,[27][28][29] but in most of the cases, the results obtained were more successful in segmenting the lumbar region. A possible solution to overcome this problem is to use an atlas-based segmentation.…”