PURPOSE:The purpose of this study is to evaluate the performance of a novel deep learning (DL) tool for fully automated measurements of the sagittal spinopelvic balance from X-ray images of the spine in comparison with manual measurements.METHODS:Ninety-seven conventional upright sagittal X-ray images from 55 subjects were retrospectively included in this study. Measurements of the parameters of the sagittal spinopelvic balance, i.e., the sacral slope (SS), pelvic tilt (PT), spinal tilt (ST), pelvic incidence (PI) and spinosacral angle (SSA), were obtained manually by identifying specific anatomical landmarks using the SurgiMap Spine software and by the fully automated DL tool. Statistical analysis was performed in terms of the mean absolute difference (MAD), standard deviation (SD) and Pearson correlation, while the paired t test was used to search for statistically significant differences between manual and automated measurements.RESULTS:The differences between reference manual measurements and those obtained automatically by the DL tool were, respectively, for SS, PT, ST, PI and SSA, equal to 5.0° (3.4°), 2.7° (2.5°), 1.2° (1.2°), 5.5° (4.2°) and 5.0° (3.5°) in terms of MAD (SD), with a statistically significant corresponding Pearson correlation of 0.73, 0.90, 0.95, 0.81 and 0.71. No statistically significant differences were observed between the two types of measurement (p value always above 0.05).CONCLUSION:The differences between measurements are in the range of the observer variability of manual measurements, indicating that the DL tool can provide clinically equivalent measurements in terms of accuracy but superior measurements in terms of cost-effectiveness, reliability and reproducibility.

Izhodišča: Sagitalna orientacija medenice je pomemben element sagitalnega ravnovesja, kvantitativno pa jo lahko opredelimo na podlagi merjenja geometrijskih parametrov medenice, in sicer naklona križnične končne ploskve (SS), nagiba medenice (PT) in naklona medenice (PI). V tem članku predstavljamo rezultate popolnoma samodejnega računalniško podprtega merjenja parametrov sagitalne orientacije medenice na podlagi rentgenskih slik ter testiramo hipotezo, da ni statistično pomembnih razlik med dobljenimi in referenčnimi ročnimi meritvami. Metode: Samodejno računalniško podprto merjenje parametrov sagitalne orientacije medenice temelji na najnovejših tehnologijah iz področja obdelave in analize medicinskih slik, in sicer na konvolucijskih nevronskih mrežah kot posebni obliki tehnik globokega učenja. Na podlagi teh tehnologij se v sagitalni rentgenski sliki medenice najprej samodejno določijo območja zanimanja (križnična končna ploskev ter kolčni sklepni glavi), nato pa se znotraj teh območij določijo značilne točke, in sicer anteriorni rob, središče in posteriorni rob križnične končne ploskve, na katere se kasneje prilega premica, ter središči obeh kolčnih sklepnih glav s pripadajočo sredinsko točko, ki predstavlja os medenice. Na podlagi osi medenice ter premice vzdolž križnične končne ploskve in njenega središča lahko končno izračunamo SS, PT in PI. Rezultati: Merjenje je bilo retrospektivno opravljeno na sagitalnih rentgenskih slikah medenice 38 oseb (15 moških in 23 žensk; povprečna starost 71,1 let). Statistična analiza referenčnih ročnih in samodejnih računalniško podprtih meritev parametrov sagitalne orientacije medenice je pokazala na relativno dobro ujemanje in majhno odstopanje. Za SS, PT in PI je bila povprečna absolutna razlika (standardni odklon) namreč 5,2º (3,8º), 2,2º (2,0º) in 5,1º (4,4º), korelacijski koeficient 0,73, 0,94 in 0,82 (p < 10-6), ničelna hipoteza pa je bila na podlagi parnega t-testa vedno potrjena (p > 0,05). Zaključek: Rezultati so pokazali, da ni statistično pomembnih razlik med referenčnimi ročnimi ter samodejnimi računalniško podprtimi meritvami parametrov sagitalne orientacije medenice. Poleg tega so odstopanja od referenčnih ročnih meritev znotraj ponovljivosti in zanesljivosti samega ročnega določanja teh parametrov, zato je z samodejnim računalniško podprtim merjenjem mogoče natančno določiti parametre sagitalne orientacije medenice. Vsekakor pa pregleda in potrjevanja tako izmerjenih vrednosti ne smemo popolnoma opustiti, saj so lahko odstopanja v določenih primerih precej velika, predvsem zaradi naravne biološke variabilnosti človeške anatomije ter lastnosti rentgenskega slikanja.

One of the most important parameters of sagittal pelvic alignment is the pelvic incidence (PI), which is commonly measured from sagittal X-ray images of the pelvis as the angle between the line connecting the midpoint of the femoral head centers with the center of the sacral endplate, and the line orthogonal to the sacral endplate. In this paper, we present the results of a fully automated measurement of PI from X-ray images that is based on the deep learning technologies. In each sagittal X-ray image of the pelvis, regions of interest (sacral endplate and both femoral heads) are first automatically defined, and then landmarks are detected within these regions, i.e. the anterior edge, the center and the posterior edge of the sacral endplate that define the line of the sacral endplate inclination, and the centers of both femoral heads with the corresponding midpoint representing the hip axis. From the hip axis, and the line along the sacral endplate and its center, PI is computed. Measurements were performed on X-ray pelvic images from 38 subjects (15 males/23 females; mean age 71.1 years), and statistical analysis of reference manual and fully automated measurements revealed a relatively good agreement, with the mean absolute difference ± standard deviation of 5.1 ± 4.4^∘ and Pearson correlation coefficient of R = 0.82 (p-value below 10^-6 ), with the paired t-test revealing no statistically significant differences (p-value above 0.05). The differences between reference manual and fully automated measurements were within the repeatability and reliability of manual measurements, indicating that PI can be accurately determined by the proposed fully automated approach.

To evaluate spinal deformities, the Cobb angle is the main diagnostic parameter that is usually measured on two-dimensional coronal radiographic (X-ray) images. In this paper, we propose a method for the evaluation of the three-dimensional (3D) Cobb angle from 3D spine mesh models with varying face-vertex density. For the upper-end and lower-end vertebra mesh models, the location of the vertebral body center and mesh faces that belong to the vertebral body surface are identified by unsupervised classification of mesh faces of the vertebral body, which serve only as training data, and subsequent supervised classification of all mesh faces. Adjacent mesh faces are then labeled with the same class, and after comparison to mesh faces in the training data, we label the mesh faces of the superior and inferior vertebral endplate. Finally, planes are fitted to the superior endplate of the upper-end vertebra and the inferior endplate of the lower-end vertebra, which define the 3D Cobb angle. The method was tested on 60 triangular mesh models of the scoliotic spine, and each mesh model was generated at 17 different face-vertex densities. For meshes with the mean face edge length below 6 mm, the proposed method was accurate, with the mean absolute error of 3.0^∘ and the corresponding standard deviation of 2.2^∘ when compared to reference measurements.

Accurate boundary delineation and segmentation of pathological spines is indispensable in spine-related applications that rely on the knowledge of vertebral shape. However, exact vertebral boundaries are often difficult to determine due to articulation of vertebrae with each other that may cause vertebral overlaps in segmentations of adjacent vertebrae. To solve this problem, we propose a novel method that consists of two steps. In the first step, the probability maps that determine vertebral boundaries are obtained from a two-way convolutional neural network, trained on normal thoracolumbar spines. In the second step, a collision-based model that consists of (at least two) consecutive vertebra mesh models is initialized close to the observed vertebrae and vertices of each mesh are displaced towards the detected boundaries. As this can lead to mesh collisions in the form of vertices of one mesh penetrating the adjacent one (and/or vice versa), these vertices are efficiently detected and then driven out of the adjacent mesh while locally preserving the shape of the corresponding mesh. By applying the proposed method to 15 three-dimensional computed tomography images of the lumbar spine containing 75 normal and fractured vertebrae, quantitative comparison against reference vertebra segmentations yielded an overall mean Dice similarity coefficient of 93.2%, mean symmetric surface distance of 0.5 mm, and Hausdorff distance of 8.4 mm.

The Cobb angle is the main diagnostic parameter for evaluating spinal deformities. Traditionally, it is measured on two-dimensional coronal radiographic (X-ray) images. In this study, we present a semi-automated algorithm for the evaluation of the Cobb angle from three-dimensional mesh models of the spine. The method was tested on 22 spine models, and the obtained mean absolute error of 2.89° against reference measurements indicates that the method performs well.

The vertebral column is a complex anatomical construct, composed of vertebrae and intervertebral discs (IVDs) supported by ligaments and muscles. During life, all components undergo degenerative changes, which may in some cases cause severe, chronic and debilitating low back pain. The main diagnostic challenge is to locate the pain generator, and degenerated IVDs have been identified to act as such. Accurate and robust segmentation of IVDs is therefore a prerequisite for computer-aided diagnosis and quantification of IVD degeneration, and can be also used for computer-assisted planning and simulation in spinal surgery. In this paper, we present a novel fully automated framework for supervised segmentation of IVDs from three-dimensional (3D) magnetic resonance (MR) spine images. By considering global intensity appearance and local shape information, a landmark-based approach is first used for the detection of IVDs in the observed image, which then initializes the segmentation of IVDs by coupling deformable models with convolutional networks (ConvNets). For this purpose, a 3D ConvNet architecture was designed that learns rich high-level appearance representations from a training repository of IVDs, and then generates spatial IVD probability maps that guide deformable models towards IVD boundaries. By applying the proposed framework to 15 3D MR spine images containing 105 IVDs, quantitative comparison of the obtained against reference IVD segmentations yielded an overall mean Dice coefficient of 92.8%, mean symmetric surface distance of 0.4 mm and Hausdorff surface distance of 3.7 mm.

Computerized segmentation of pathological structures in medical images is challenging, as, in addition to unclear image boundaries, image artifacts, and traces of surgical activities, the shape of pathological structures may be very different from the shape of normal structures. Even if a sufficient number of pathological training samples are collected, statistical shape modeling cannot always capture shape features of pathological samples as they may be suppressed by shape features of a considerably larger number of healthy samples. At the same time, landmarking can be efficient in analyzing pathological structures but often lacks robustness. In this paper, we combine the advantages of landmark detection and deformable models into a novel supervised multi-energy segmentation framework that can efficiently segment structures with pathological shape. The framework adopts the theory of Laplacian shape editing, that was introduced in the field of computer graphics, so that the limitations of statistical shape modeling are avoided. The performance of the proposed framework was validated by segmenting fractured lumbar vertebrae from 3-D computed tomography images, atrophic corpora callosa from 2-D magnetic resonance (MR) cross-sections and cancerous prostates from 3D MR images, resulting respectively in a Dice coefficient of 84.7 ± 5.0%, 85.3 ± 4.8% and 78.3 ± 5.1%, and boundary distance of 1.14 ± 0.49mm, 1.42 ± 0.45mm and 2.27 ± 0.52mm. The obtained results were shown to be superior in comparison to existing deformable model-based segmentation algorithms.

Adolescent idiopathic scoliosis (AIS) is a 3-D deformation of the spine. Identifying curve progression in AIS at the first visit is a clinically relevant problem but remains challenging due to lack of relevant descriptors. We present here a classification framework to identify patients whose spine deformity will progress from those who will remain stable. The method uses personalized 3-D spine reconstructions at baseline from progressive (P) and non-progressive (NP) patients to train a predictive model. Morphological changes between groups are detected using a manifold learning algorithm based on Grassmannian kernels in order to assess the similarity between shape topology and inter-vertebral poses in both groups (P, NP). We test the method to classify 52 progressive and 81 non-progressive patients enrolled in a prospective clinical study, yielding classification rates comparing favorably to standard classification methods.

Gradual degeneration of intervertebral discs of the lumbar spine is one of the most common causes of low back pain. A fully automatic, accurate and robust segmentation of intervertebral discs in magnetic resonance (MR) images is therefore a prerequisite for the computer-aided diagnosis and quantification of intervertebral disc degeneration. In this paper, we propose an automated framework for intervertebral disc segmentation from MR spine images, in which intervertebral disc detection is performed by a landmark-based approach and segmentation by a deformable model-based approach using the self-similarity context (SSC) descriptor. The performance was evaluated on three publicly available databases of MR spine images that represent the training, on-line and on-site testing data for the intervertebral disc localization and segmentation challenge in conjunction with the 3rd MICCAI Workshop Challenge on Computational Methods and Clinical Applications for Spine Imaging - MICCAI–CSI2015, yielding an overall mean Euclidean distance of 2.4, 1.7 and 2.2 mm for intervertebral disc localization, and an overall mean Dice coefficient of 92.5, 91.5 and 92.0

A multiple center milestone study of clinical vertebra segmentation is presented in this paper. Vertebra segmentation is a fundamental step for spinal image analysis and intervention. The first half of the study was conducted in the spine segmentation challenge in 2014 International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) Workshop on Computational Spine Imaging (CSI 2014). The objective was to evaluate the performance of several state-of-the-art vertebra segmentation algorithms on computed tomography (CT) scans using ten training and five testing dataset, all healthy cases; the second half of the study was conducted after the challenge, where additional 5 abnormal cases are used for testing to evaluate the performance under abnormal cases. Dice coefficients and absolute surface distances were used as evaluation metrics. Segmentation of each vertebra as a single geometric unit, as well as separate segmentation of vertebra substructures, was evaluated. Five teams participated in the comparative study. The top performers in the study achieved Dice coefficient of 0.93 in the upper thoracic, 0.95 in the lower thoracic and 0.96 in the lumbar spine for healthy cases, and 0.88 in the upper thoracic, 0.89 in the lower thoracic and 0.92 in the lumbar spine for osteoporotic and fractured cases. The strengths and weaknesses of each method as well as future suggestion for improvement are discussed. This is the first multi-center comparative study for vertebra segmentation methods, which will provide an up-to-date performance milestone for the fast growing spinal image analysis and intervention.

Adolescent idiopathic scoliosis (AIS) is a complex three-dimensional (3-D) deformation of the trunk, which includes lateral deviation of the spine, asymmetric deformation and axial rotation of the vertebrae, deformation of the rib cage, and possibly of the pelvis. In order to analyze 3-D characteristics of spinal deformations related to AIS, we propose a coarse-to-fine 3-D modeling framework of the scoliotic spine from biplanar X-ray images. First, the spine centerline represented by a cubic spline is used as a data descriptor for the underlying sparse modeling approach to obtain an initial 3-D reconstruction. We then optimize a multi-object pose+shape model that uses the statistical pose and geometric shape variability from a training set of scoliotic spines, onto a set of adjusted spinal landmarks to obtain the final 3-D reconstruction. The performance was evaluated on a database of 844 AIS patients, yielding an overall mean root-mean-square Euclidean distance 2.48 ± 0.68 mm for the final 3-D reconstruction of the scoliotic spine.

Automated detection and segmentation of vertebral bodies from spinal computed tomography (CT) images is usually a prerequisite step for numerous spine-related medical applications, such as diagnosis, surgical planning and follow-up assessment of spinal pathologies. However, automated detection and segmentation are challenging tasks due to a relatively high degree of anatomical complexity, presence of unclear boundaries and articulation of vertebrae with each other. In this paper, we describe a sparse representation error minimization (SEM) framework for joint detection and segmentation of vertebral bodies in CT images. By minimizing the sparse representation error of sampled intensity values, we are able to recover the oriented bounding box (OBB) and segmentation binary mask for each vertebral body in the CT image. The performance of the proposed SEM framework was evaluated on five CT images of the thoracolumbar spine. The resulting Euclidean distance of 1:75±1:02 mm, computed between the center points of recovered and corresponding reference OBBs, and Dice coefficient of 92:3±2:7%, computed between the resulting and corresponding reference segmentation binary masks, indicate that the proposed framework can successfully detect and segment vertebral bodies in CT images of the thoracolumbar spine.

We propose an automated method for supervised segmentation of vertebral bodies (VBs) from three-dimensional (3D) magnetic resonance (MR) spine images that is based on coupling deformable models with convolutional neural networks (CNNs). We designed a 3D CNN architecture that learns the appearance from a training set of VBs to generate 3D spatial VB probability maps, which guide deformable models towards VB boundaries. The proposed method was applied to segment 161 VBs from 3D MR spine images of 23 subjects, and the results were compared to reference segmentations. By yielding an overall Dice similarity coefficient of 93.4 ± 1.7 % , mean symmetric surface distance of 0.54 ± 0.14 mm and Hausdorff distance of 3.83 ± 1.04 mm , the proposed method proved superior to existing VB segmentation methods.

The evaluation of changes in Intervertebral Discs (IVDs) with 3D Magnetic Resonance (MR) Imaging (MRI) can be of interest for many clinical applications. This paper presents the evaluation of both IVD localization and IVD segmentation methods submitted to the Automatic 3D MRI IVD Localization and Segmentation challenge, held at the 2015 International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI2015) with an on-site competition. With the construction of a manually annotated reference data set composed of 25 3D T2-weighted MR images acquired from two different studies and the establishment of a standard validation framework, quantitative evaluation was performed to compare the results of methods submitted to the challenge. Experimental results show that overall the best localization method achieves a mean localization distance of 0.8 mm and the best segmentation method achieves a mean Dice of 91.8%, a mean average absolute distance of 1.1 mm and a mean Hausdorff distance of 4.3 mm, respectively. The strengths and drawbacks of each method are discussed, which provides insights into the performance of different IVD localization and segmentation methods.

Automated and semi-automated detection and segmentation of spinal and vertebral structures from computed tomography (CT) images is a challenging task due to a relatively high degree of anatomical complexity, presence of unclear boundaries and articulation of vertebrae with each other, as well as due to insufficient image spatial resolution, partial volume effects, presence of image artifacts, intensity variations and low signal-to-noise ratio. In this paper, we describe a novel framework for automated spine and vertebrae detection and segmentation from 3-D CT images. A novel optimization technique based on interpolation theory is applied to detect the location of the whole spine in the 3-D image and, using the obtained location of the whole spine, to further detect the location of individual vertebrae within the spinal column. The obtained vertebra detection results represent a robust and accurate initialization for the subsequent segmentation of individual vertebrae, which is performed by an improved shape-constrained deformable model approach. The framework was evaluated on two publicly available CT spine image databases of 50 lumbar and 170 thoracolumbar vertebrae. Quantitative comparison against corresponding reference vertebra segmentations yielded an overall mean centroid-to-centroid distance of 1.1 mm and Dice coefficient of 83.6% for vertebra detection, and an overall mean symmetric surface distance of 0.3 mm and Dice coefficient of 94.6% for vertebra segmentation. The results indicate that by applying the proposed automated detection and segmentation framework, vertebrae can be successfully detected and accurately segmented in 3-D from CT spine images.

Detection of an object of interest can be represented as an optimization problem that can be solved by brute force or heuristic algorithms. However, the globally optimal solution may not represent the optimal detection result, which can be especially observed in the case of vertebra detection, where neighboring vertebrae are of similar appearance and shape. An adequate optimizer has to therefore consider not only the global optimum but also local optima that represent candidate locations for each vertebra. In this paper, we describe a novel framework for automated spine and vertebra detection in three-dimensional (3D) images of the lumbar spine, where we apply a novel optimization technique based on interpolation theory to detect the location of the whole spine in the 3D image and to detect the location of individual vertebrae within the spinal column. The performance of the proposed framework was evaluated on $$10$$ computed tomography (CT) images of the lumbar spine. The resulting mean symmetric absolute surface distance of $$1.25\,{\pm }\,0.41$$ mm and Dice coefficient of $$83.67\,{\pm }\,4.44$$ %, computed from the final vertebra detection results against corresponding reference vertebra segmentations, indicate that the proposed framework can successfully detected vertebrae in CT images of the lumbar spine.

This paper presents a method for automatic vertebra segmentation. The method consists of two parts: vertebra detection and vertebra segmentation. To detect vertebrae in an unknown CT spine image, an interpolation-based optimization approach is first applied to detect the whole spine, then to detect the location of individual vertebrae, and finally to rigidly align shape models of individual vertebrae to the detected vertebrae. Each optimization is performed using a spline-based interpolation function on an equidistant sparse optimization grid to obtain the optimal combination of translation, scaling and/or rotation parameters. The computational complexity in examining the parameter space is reduced by a dimension-wise algorithm that iteratively takes into account only a subset of parameter space dimensions at the time. The obtained vertebra detection results represent a robust and accurate initialization for the subsequent segmentation of individual vertebrae, which is built upon the existing shape-constrained deformable model approach. The proposed iterative segmentation consists of two steps that are executed in each iteration. To find adequate boundaries that are distinctive for the observed vertebra, the boundary detection step applies an improved robust and accurate boundary detection using Canny edge operator and random forest regression model that incorporates prior knowledge through image intensities and intensity gradients. The mesh deformation step attracts the mesh of the vertebra shape model to vertebra boundaries and penalizes the deviations of the mesh from the training repository while preserving shape topology.

STUDY DESIGN:Pilot single-centre, stratified, prospective, randomized, double-blinded, parallel-group, controlled study.OBJECTIVE:To determine whether vertebral end-plate perforation after lumbar discectomy causes annulus reparation and intervertebral disc volume restoration. To determine that after 6 months there would be no clinical differences between the control and study group.SUMMARY OF BACKGROUND DATA:Low back pain is the most common long-term complication after lumbar discectomy. It is mainly caused by intervertebral disc space loss, which promotes progressive degeneration. This is the first study to test the efficiency of a previously described method (vertebral end-plate perforation) that should advocate for annulus fibrosus reparation and disc space restoration.METHODS:We selected 30 eligible patients according to inclusion and exclusion criteria and randomly assigned them to the control (no end-plate perforation) or study (end-plate perforation) group. Each patient was evaluated in 5 different periods, where data were collected [preoperative and 6-mo follow-up magnetic resonance imaging and functional outcome data: visual analogue scale (VAS) back, VAS legs, Oswestry disability index (ODI)]. Intervertebral space volume (ISV) and height (ISH) were measured form the magnetic resonance images. Statistical analysis was performed using paired t test and linear regression. P<0.05 was considered statistically significant.RESULTS:We found no statistically significant difference between the control group and the study group concerning ISV (P=0.6808) and ISH (P=0.8981) 6 months after surgery. No statistically significant differences were found between ODI, VAS back, and VAS legs after 6 months between the 2 groups, however, there were statistically significant differences between these parameters in different time periods. Correlation between the volume of disc tissue removed and preoperative versus postoperative difference in ISV was statistically significant (P=0.0020).CONCLUSIONS:The present study showed positive correlation between the volume of removed disc tissue and decrease in postoperative ISV and ISH. There were no statistically significant differences in ISV and ISH between the group with end-plate perforation and the control group 6 months after lumbar discectomy. Clinical outcome and disability were significantly improved in both groups 3 and 6 months after surgery.