Luc Duong

Three-Dimensional Classification of Spinal Deformities Using Fuzzy Clustering

Study Design. A prospective study of a large set of three-dimensional (3D) reconstructions of spinal deformities in adolescent idiopathic scoliosis (AIS). Objectives. To determine the value of fuzzy clustering techniques to automatically detect clinically relevant 3D curve patterns within this set of 3D spine models. Summary of Background Data. Classification is important for the assessment of AIS and has been mainly used to guide surgical treatment. Current classification systems are based on visual curve pattern identification using two-dimensional radiologic measurements but remain controversial because of their low interobserver and intraobserver reliability. A clinically useful 3D classification remains to be found. Methods. An unsupervised learning algorithm, fuzzy k-means clustering, was applied on 409 3D spine models. Analysis of data distribution using clinical parameters was performed by studying similar curve patterns, near each cluster center identified. Results. The algorithm determined that the entire sample of models could be segmented in five easily differentiated curve patterns similar to those of the Lenke and King classifications. Furthermore, a system with 12 classes made possible the identification of subpatterns of spinal deformity with true 3D components. Conclusions. Automatic and clinically relevant 3D classification of AIS is possible using an unsupervised learning algorithm. This approach can now be used to build a relevant 3D classification of AIS using appropriate key features of 3D models selected by a panel of expert spinal deformity surgeons.

Reliability of the Spinal Deformity Study Group classification of lumbosacral spondylolisthesis.

Study Design. Reliability study of the computer-assisted SDSG (Spinal Deformity Study Group) classification of lumbosacral spondylolisthesis. Objective. To assess the intra- and interobserver reliability of the computer-assisted SDSG classification of lumbosacral spondylolisthesis. Summary of Background Data. The SDSG has proposed a new classification of lumbosacral spondylolisthesis based on slip grade, pelvic incidence (PI), and sacro-pelvic and spinal balance. Three types of low-grade spondylolisthesis are described: low PI (type 1), normal PI (type 2), and high PI (type 3). High-grade spondylolisthesis are defined as type 4 (balanced sacro-pelvis), type 5 (retroverted sacro-pelvis with balanced spine), and type 6 (retroverted sacro-pelvis with unbalanced spine). Methods. Full-length standing lateral radiographs of the spine of 40 subjects with lumbosacral spondylolisthesis were reviewed twice by 7 observers. Custom software was used by the observers to identify 7 anatomical landmarks on each radiograph to determine the SDSG type for all subjects. Percentage of agreement and &kgr; coefficients were used to determine the intra- and interobserver reliability. Results. All 6 types of spondylolisthesis described in the computer-assisted SDSG classification were identified. Overall intra- and interobserver agreements were 80% (&kgr;: 0.74) and 71% (&kgr;: 0.65), respectively. The intra- and interobserver agreements associated with computerized determination of slip grade were 92% (&kgr;: 0.83) and 88% (&kgr;: 0.78), respectively. As for computerized determination of sacro-pelvic and spinal balance, intra- and interobserver agreements were 86% (&kgr;: 0.76) and 75% (&kgr;: 0.63) for low-grade slips, whereas they were 88% (&kgr;: 0.80) and 83% (&kgr;: 0.75) for high-grade slips. Conclusion. Substantial intra- and interobserver reliability was found for the computer-assisted SDSG classification, and all 6 types of lumbosacral spondylolisthesis were identified. Refinement of the computer-assisted classification technique is, however, needed to further increase the reliability of the SDSG classification and facilitate its clinical use.

Three-dimensional subclassification of Lenke type 1 scoliotic curves.

Study Design Prospective study of the 3-D shape variability of spinal curve in Lenke type 1 adolescent idiopathic scoliosis (AIS). Objectives To determine the statistical 3-D variability of Lenke type 1 curves and to evaluate clinical parameters that can be integrated to refine the Lenke et al original proposal, and to pave the road for a comprehensive 3-D subclassification of AIS. Summary of Background Data Several classification systems based on the identification of key features from frontal and sagittal x-rays have been proposed in AIS, but these remain an oversimplification of the complex 3-D deformity because it is only based on 2-D imaging. Clinical 3-D parameter variability has been investigated in previous studies, but has never been considered in the context of the Lenke classification. Methods Radiographs of 68 AIS patients with Lenke type 1 curves were reconstructed in 3-dimension using a stereo-radiographic technique and were submitted to a computer algorithm to compute a set of 3-D parameters that can be used to characterize the 3-D curve. Cluster analysis was performed to determine the statistical distribution of 3-D parameters among Lenke 1 curve types. Results Statistical analysis shows specific 3-D deformation patterns within Lenke type 1 curves, mostly using the best-fit plane or BFP (SD±22.9, ±49.8) and geometric torsion parameters. No significant variability was found using the plane of maximum curvature or PMC. Conclusions Recent advances in computer vision facilitate the introduction of 3-D reconstruction in a standard clinical setting and can provide more information toward the spine behavior in 3-D space. A direct consequence of commonly used 3-D reconstruction would be to be able to evaluate 3-D indices and to devise a real 3-D classification system from the Lenke et al proposal.

Vesselwalker: Coronary arteries segmentation using random walks and hessian-based vesselness filter

The segmentation of vascular structures from 2D X-ray angiographies is an important step for vessel measurement, diagnosis and treatment planning. Segmentation of such structures can be challenging due to the vessel appearance and topology. In this paper, we propose a novel interactive method to segment vascular structures by combining Hessian-based vesselness information and the random walk formulation, in which manually selected seed points can be used to refine the segmentation result. The proposed method was tested on coronary arteries angiograms and has shown to be more accurate than an active contour-based method or the Random Walker algorithm, with a mean AUC of 97.2%.

Curve-based 2D-3D registration of coronary vessels for image guided procedure

3D roadmap provided by pre-operative volumetric data that is aligned with fluoroscopy helps visualization and navigation in Interventional Cardiology (IC), especially when contrast agent-injection used to highlight coronary vessels cannot be systematically used during the whole procedure, or when there is low visibility in fluoroscopy for partially or totally occluded vessels. The main contribution of this work is to register pre-operative volumetric data with intraoperative fluoroscopy for specific vessel(s) occurring during the procedure, even without contrast agent injection, to provide a useful 3D roadmap. In addition, this study incorporates automatic ECG gating for cardiac motion. Respiratory motion is identified by rigid body registration of the vessels. The coronary vessels are first segmented from a multislice computed tomography (MSCT) volume and correspondent vessel segments are identified on a single gated 2D fluoroscopic frame. Registration can be explicitly constrained using one or multiple branches of a contrast-enhanced vessel tree or the outline of guide wire used to navigate during the procedure. Finally, the alignment problem is solved by Iterative Closest Point (ICP) algorithm. To be computationally efficient, a distance transform is computed from the 2D identification of each vessel such that distance is zero on the centerline of the vessel and increases away from the centerline. Quantitative results were obtained by comparing the registration of random poses and a ground truth alignment for 5 datasets. We conclude that the proposed method is promising for accurate 2D-3D registration, even for difficult cases of occluded vessel without injection of contrast agent.

Deep feature learning for automatic tissue classification of coronary artery using optical coherence tomography

Kawasaki disease (KD) is an acute childhood disease complicated by coronary artery aneurysms, intima thickening, thrombi, stenosis, lamellar calcifications, and disappearance of the media border. Automatic classification of the coronary artery layers (intima, media, and scar features) is important for analyzing optical coherence tomography (OCT) images recorded in pediatric patients. OCT has been known as an intracoronary imaging modality using near-infrared light which has recently been used to image the inner coronary artery tissues of pediatric patients, providing high spatial resolution (ranging from 10 to 20 μm). This study aims to develop a robust and fully automated tissue classification method by using the convolutional neural networks (CNNs) as feature extractor and comparing the predictions of three state-of-the-art classifiers, CNN, random forest (RF), and support vector machine (SVM). The results show the robustness of CNN as the feature extractor and random forest as the classifier with classification rate up to 96%, especially to characterize the second layer of coronary arteries (media), which is a very thin layer and it is challenging to be recognized and specified from other tissues.

Assessment of sacral doming in lumbosacral spondylolisthesis.

Study Design. Quantitative versus subjective evaluation of sacral doming in lumbosacral spondylolisthesis. Objectives. To evaluate the relevance of the Spinal Deformity Study Group (SDSG) index in the assessment of sacral doming and to propose a quantitative criterion to differentiate between significant and nonsignificant doming. Summary of Background Data. There is no consensus on the optimal technique to assess sacral doming, although it is an important feature in spondylolisthesis. Methods. Five spinal surgeons subjectively assessed the sacral endplate of 100 subjects (34 high-grade spondylolisthesis, 50 low-grade spondylolisthesis, 16 controls) from lateral radiographs. Subjects were classified by each surgeon as having significant or nonsignificant sacral doming. An independent observer quantitatively evaluated sacral doming for all subjects using the SDSG index. A criterion to differentiate significant from nonsignificant sacral doming was sought, based on the comparison between the subjective assessment of surgeons and the quantitative evaluation by the independent observer. Intrarater and interrater reliability of the SDSG index was evaluated using intraclass correlation coefficient (ICC). Results. Intrarater and interrater ICCs for the SDSG index were excellent at 0.91 and 0.88, respectively. Sacral doming evaluated with the SDSG index was 11.6% ± 5.0% (range, 1.5%–18.9%), 16.4% ± 6.3% (range, 3.7%–35.6%), and 27.9% ± 10.9% (range, 5.7%–56.9%) for controls, low-grade, and high-grade cases, respectively. Overall intersurgeon agreement on the significance of sacral doming was substantial at 88% (kappa = 0.72). With a threshold value of 25% for the SDSG index, 93% of concordance was found between the quantitative evaluation using the SDSG index and the multisurgeons subjective assessment. Conclusion. This study confirms the relevance of the SDSG index to assess sacral doming in lumbosacral spondylolisthesis. The authors propose a criterion of 25% to differentiate significant from nonsignificant sacral doming using the SDSG index. Such a criterion will allow more accurate assessment of sacral remodeling, especially for borderline cases, and facilitate comparisons between studies.

Real time noninvasive assessment of external trunk geometry during surgical correction of adolescent idiopathic scoliosis

BackgroundThe correction of trunk deformity is crucial in scoliosis surgery, especially for the patients self-image. However, direct visualization of external scoliotic trunk deformity during surgical correction is difficult due to the covering draping sheets.MethodsAn optoelectronic camera system with 10 passive markers is used to track the trunk geometry of 5 scoliotic patients during corrective surgery. The position of 10 anatomical landmarks and 5 trunk indices computed from the position of the passive markers are compared during and after instrumentation of the spine.ResultsInternal validation of the accuracy of tracking was evaluated at 0.41 +/- 0.05 mm RMS. Intra operative tracking during surgical maneuvers shows improvement of the shoulder balance during and after correction of the spine. Improvement of the overall patient balance is observed. At last, a minor increase of the spinal length can be noticed.ConclusionTracking of the external geometry of the trunk during surgical correction is useful to monitor changes occurring under the sterile draping sheets. Moreover, this technique can used be used to reach the optimal configuration on the operating frame before proceeding to surgery. The current tracking technique was able to detect significant changes in trunk geometry caused by posterior instrumentation of the spine despite significant correction of the spinal curvature. It could therefore become relevant for computer-assisted guidance of surgical maneuvers when performing posterior instrumentation of the scoliotic spine, provide important insights during positioning of patients.

Automatic Detection of Scoliotic Curves in Posteroanterior Radiographs

Spinal deformities are diagnosed using posteroanterior (PA) radiographs. Automatic detection of the spine on conventional radiographs would be of interest to quantify curve severity, would help reduce observer variability and would allow large-scale retrospective studies on radiographic databases. The goal of this paper is to present a new method for automatic detection of spinal curves from a PA radiograph. A region of interest (ROI) is first extracted according to the 2-D shape variability of the spine obtained from a set of PA radiographs of scoliotic patients. This region includes 17 bounding boxes delimiting each vertebral level from T1 to L5. An adaptive filter combining shock with complex diffusion is used to individually restore the image of each vertebral level. Then, texture descriptors of small block elements are computed and submitted for training to support vector machines (SVM). Vertebral bodys locations are thereby inferred for a particular vertebral level. The classifications of block elements for all 17 SVMs are identified in the image and a voting system is introduced to cumulate correctly predicted blocks. A spline curve is then fitted through the centers of the predicted vertebral regions and compared to a manual identification using a Student t-test. A clinical validation is performed using 100 radiographs of scoliotic patients (not used for training) and the detected spinal curve is found to be statistically similar (p < 0.05) in 93% of cases to the manually identified curve.

Interobserver and Intraobserver Variability in the Identification of the Lenke Classification Lumbar Modifier in Adolescent Idiopathic Scoliosis

Study Design Interobserver and intraobserver reliability study for the identification of the Lenke classification lumbar modifier by a panel of experts compared with a computer algorithm. Objectives To measure the variability of the Lenke classification lumbar modifier and determine if computer assistance using 3-dimensional spine models can improve the reliability of classification. Summary of Background Data The lumbar modifier has been proposed to subclassify Lenke scoliotic curve types into A, B, and C on the basis of the relationship between the central sacral vertical line (CSVL) and the apical lumbar vertebra. Landmarks for identification of the CSVL have not been clearly defined, and the reliability of the actual CSVL position and lumbar modifier selection have never been tested independently. Therefore, the value of the lumbar modifier for curve classification remains unknown. Methods The preoperative radiographs of 68 patients with adolescent idiopathic scoliosis presenting a Lenke type 1 curve were measured manually twice by 6 members of the Scoliosis Research Society 3-dimensional classification committee at 6 months interval. Intraobserver and interobserver reliability was quantified using the percentage of agreement and κ statistics. In addition, the lumbar curve of all subjects was reconstructed in 3-dimension using a stereoradiographic technique and was submitted to a computer algorithm to infer the lumbar modifier according to measurements from the pedicles. Results Interobserver rates for the first trial showed a mean κ value of 0.56. Second trial rates were higher with a mean κ value of 0.64. Intraobserver rates were evaluated at a mean κ value of 0.69. The computer algorithm was successful in identifying the lumbar curve type and was in agreement with the observers by a proportion up to 93%. Conclusions Agreement between and within observers for the Lenke lumbar modifier is only moderate to substantial with manual methods. Computer assistance with 3-dimensional models of the spine has the potential to decrease this variability.