Computed Tomography Image Volumes Research Articles

An attention-enhanced cross-task network to analyse lung nodule attributes in CT images

Accurate characterization of visual attributes such as spiculation, lobulation, and calcification of lung nodules in computed tomography (CT) images is critical in cancer management. The characterization of these attributes is often subjective, which may lead to high inter- and intra-observer variability. Furthermore, lung nodules are often heterogeneous in the cross-sectional image slices of a 3D volume. Current state-of-the-art methods that score multiple attributes rely on deep learning-based multi-task learning (MTL) schemes. These methods, however, extract shared visual features across attributes and then examine each attribute without explicitly leveraging their inherent intercorrelations. Furthermore, current methods treat each slice with equal importance without considering their relevance or heterogeneity, which limits performance. In this study, we address these challenges with a new convolutional neural network (CNN)-based MTL model that incorporates multiple attention-based learning modules to simultaneously score 9 visual attributes of lung nodules in CT image volumes. Our model processes entire nodule volumes of arbitrary depth and uses a slice attention module to filter out irrelevant slices. We also introduce cross-attribute and attribute specialization attention modules that learn an optimal amalgamation of meaningful representations to leverage relationships between attributes. We demonstrate that our model outperforms previous state-of-the-art methods at scoring attributes using the well-known public LIDC-IDRI dataset of pulmonary nodules from over 1,000 patients. Our model also performs competitively when repurposed for benign-malignant classification. Our attention modules provide easy-to-interpret weights that offer insights into the predictions of the model.

Usefulness of Preoperative Planning by Three-Dimensional Planning Software for Pedicle Screw Placement in Thoracolumbar Surgeries: Misplacement Rate and Associated Risk Factors.

Introduction A number of imaging technologies have been developed to reduce the risk of pedicle screw (PS) misplacement. For example, preoperative three-dimensional (3D) planning can reportedly enhance implant placement accuracy in some orthopedic surgeries. However, no study has investigated the effect of preoperative 3D planning on PS placement without intraoperative 3D navigation. Thus, in this study, we aim to examine the accuracy of PS placement and identify the risk factors for PS misplacement in thoracolumbar surgeries performed using preoperative 3D planning software with intraoperative fluoroscopic guidance in a retrospective study. Methods In total, 25 consecutive patients (197 PSs) underwent thoracic or lumbar spinal fusion surgeries using preoperative 3D planning with intraoperative fluoroscopic guidance. PS misplacement was graded based on the degree of perforation (Grade 0, no perforation; Grade 1, <2 mm; Grade 2, 2-4 mm; Grade 3, >4 mm) observed in postoperative computed tomography (CT). Deviations between planned and actual PSs were evaluated by matching preoperative and postoperative CT volume images for each vertebra. Results The overall PS misplacement rate was 6.6% (Grade 1: 4.0%, Grade 2: 1.5%, Grade 3: 1.0%). The median linear deviations of PS entry points between planned and actual locations were determined to be 3.3 mm and 3.3 mm for the horizontal and vertical axes, respectively. The median angular deviations of the PS axis were 6.2° and 4.5° for the transverse and sagittal planes, respectively. Multivariate analysis revealed that horizontal deviation of the PS entry point was the sole factor associated with Grade ≥1 PS misplacement (odds ratio=2.47, p<0.001). Conclusions Preoperative 3D planning software without intraoperative 3D navigation was able to achieve a relatively low PS misplacement ratio among the reported ratio of conventional techniques without navigation. Surgeons should carefully ensure that the entry point is consistent with preoperative planning, especially in the mediolateral direction to avoid misplacement in this method.

In-Line Phase Contrast Mammography, Phase Contrast Digital Breast Tomosynthesis, and Phase Contrast Breast Computed Tomography With a Dedicated CT Scanner and a Microfocus X-Ray Tube: Experimental Phantom Study

We performed laboratory investigations of the image quality for 2-D [digital mammography (DM)] and 3-D [digital breast tomosynthesis (DBT) and computed tomography (CT)] in-line phase contrast imaging dedicated to the breast. We employed the cone-beam, microfocus CT scanner developed in-house. Planar imaging tests were carried out with BR50/50 and polymethylmethacrylate test objects, at tube voltages of 60 kV, 80 kV, and 100 kV. CT volume images, acquired at 80 kV at dose levels comparable to two-view mammography (mean glandular dose <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\cong ~4$ </tex-math></inline-formula> mGy), as well as at 120 kV, were reconstructed from phase contrast projection images as well as from the corresponding phase retrieved projection images (Paganin’s algorithm). We assumed 3-D spatial uniform distribution of the ratio <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\delta /\beta $ </tex-math></inline-formula> of the X-ray optical coefficients of the sample material, at the effective energy of the X-ray beam. In 2-D imaging, all phantom details showed higher contrast after phase retrieval. In particular, contrast-to-noise ratio between a simulated mass and phantom background in phase retrieved images resulted up to 6.5 times higher than for absorption-based images. CT and DBT phase retrieval images of a homogeneous 50/50 test object at 120 kV produced higher signal-to-noise ratio of microcalcifications than in phase contrast DBT and CT of the same object, at the same glandular dose. In CT reconstructed volume images, microcalcifications down to 0.230–0.290-mm size presented an increased visibility after phase retrieval, in a two-components 50/50 breast phantom. For the first time, the advantages of in-line phase imaging in microfocus breast imaging have been shown with one setup for all techniques (DM, DBT, and breast CT). These results motivate further research on in-line phase contrast 2-D and 3-D imaging of the breast with a microfocus X-ray tube for potential application in the clinic of in-line phase contrast DM, DBT, and breast CT at dose levels comparable with corresponding attenuation-based exams.

Augmented reality visualisation for orthopaedic surgical guidance with pre‐ and intra‐operative multimodal image data fusion

Augmented reality (AR) has proven to be a useful, exciting technology in several areas of healthcare. AR may especially enhance the operator's experience in minimally invasive surgical applications by providing more intuitive and naturally immersive visualisation in those procedures which heavily rely on three-dimensional (3D) imaging data. Benefits include improved operator ergonomics, reduced fatigue, and simplified hand–eye coordination. Head-mounted AR displays may hold great potential for enhancing surgical navigation given their compactness and intuitiveness of use. In this work, the authors propose a method that can intra-operatively locate bone structures using tracked ultrasound (US), registers to the corresponding pre-operative computed tomography (CT) data and generates 3D AR visualisation of the operated surgical scene through a head-mounted display. The proposed method deploys optically-tracked US, bone surface segmentation from the US and CT image volumes, and multimodal volume registration to align pre-operative to the corresponding intra-operative data. The enhanced surgical scene is then visualised in an AR framework using a HoloLens. They demonstrate the method's utility using a foam pelvis phantom and quantitatively assess accuracy by comparing the locations of fiducial markers in the real and virtual spaces, yielding root mean square errors of 3.22, 22.46, and 28.30 mm in the x , y , and z directions, respectively.

Attenuation correction of SPECT images based on separately performed CT: Effect on the measurement of regional uptake values.

A new software approach uses separately acquired CT images for attenuation correction after retrospective fusion with the SPECT data. This study evaluates the effect of this CT-based attenuation correction on indium-111-pentetreotide-SPECT images. Indium-111-pentetreotide-SPECT imaging using a dual-head gamma camera e.cam (Siemens Medical Solutions, Erlangen, Germany) as well as separate spiral computed tomography (CT) was performed in 13 patients. After fusion of SPECT and CT data, the bilinear attenuation coefficients were calculated for each pixel in the CT image volume using their Hounsfield unit values and attenuation-corrected images were reconstructed iteratively (OSEM 2D). Regions of interest (ROIs) were drawn on 24 suspicious foci and background, and target to background ratios were calculated for corrected (TBAC) and uncorrected (TBNAC) images. The shortest distance from the centre of the lesion to the surface of the body (DS) was measured on the corresponding CT slice. Furthermore, ROIs were drawn over the rim and the centre of the liver. Ratios of hepatic count rates for corrected (LRAC) and uncorrected (LRNAC) images were also compared. In lesions located more centrally, TBAC was up to 52% higher, whereas in peripherally located lesions, TBAC was up to 63% lower than TBNAC. The TBAC/TBNAC quotient was linearly correlated with DS. In the liver, attenuation correction resulted in a 35% increase of LRAC compared with LRNAC. Attenuation correction of SPECT images performed by separately acquired CT data is quick and simple. It improves the contrast between target and background for lesions located more centrally in the body and improves homogeneity of the visualisation of tracer uptake in the liver.

3D segmentation of liver and its lesions using optimized geometric contours

An optimised accurate segmentation of liver and hepatic lesions from Computed Tomography (CT) scans can support the study of deriving quantitative biomarkers for accurate clinical diagnosis and successful conclusion of computer-assisted diagnosis and therapy. In spite of several existence of research, fully automatic segmentation of liver and its lesions still remains a challenging task. This paper presents a method for delineation of liver and its lesions in CT volume images using energy optimised geometric active contours. The abdominal region was initially determined based on maximum intensity projection (MIP) and thresholding methods to extract the liver volume of interest (VOI). Two geometric contours using level set method were initialized and cascaded for a combined segmentation of the liver and its lesions. In the first step, a contour to segment the liver was initialized that served as VOI input for a second geometric contour. The second contour exclusively segmented lesions within the predicted liver VOIs of first step. Experiments on CT image data obtained dice coefficient of 0.900, jaccard coefficient of 0.818 and other surface distance measures demonstrate the advantages of the proposed methodology in both exactness and competence.

The approximate entropy concept extended to three dimensions for calibrated, single parameter structural complexity interrogation of volumetric images

Reconstructive volumetric imaging permeates medical practice because of its apparently clear depiction of anatomy. However, the tell tale signs of abnormality and its delineation for treatment demand experts work at the threshold of visibility for hints of structure. Hitherto, a suitable assistive metric that chimes with clinical experience has been absent. This paper develops the complexity measure approximate entropy (ApEn) from its 1D physiological origin into a three-dimensional (3D) algorithm to fill this gap. The first 3D algorithm for this is presented in detail. Validation results for known test arrays are followed by a comparison of fan-beam and cone-beam x-ray computed tomography image volumes used in image guided radiotherapy for cancer. Results show the structural detail down to individual voxel level, the strength of which is calibrated by the ApEn process itself. The potential for application in machine assisted manual interaction and automated image processing and interrogation, including radiomics associated with predictive outcome modeling, is discussed.

Dynamic Volume Computed Tomography Imaging of the Upper Airway in Obstructive Sleep Apnea.

To describe a dynamic three-dimensional (3D) computed tomography (CT) technique for the upper airway and compare the required radiation dose to that used for common clinical studies of a similar anatomical area, such as for subjects undergoing routine clinical facial CT. Dynamic upper-airway CT was performed on eight subjects with persistent obstructive sleep apnea, four of whom were undergoing magnetic resonance imaging and an additional four subjects who had a contraindication to magnetic resonance imaging. This Health Insurance Portability and Accountability Act-compliant study was approved by our institutional review board, and informed consent was obtained. The control subjects (n = 41) for comparison of radiation dose were obtained from a retrospective review of the clinical picture-archiving computer system to identify 10 age-matched patients per age-based control group undergoing facial CT. Dynamic 3D CT can be performed with an effective radiation dose of less than 0.38 mSv, a dose that is less than or comparable to that used for clinical facial CT. The resulting data- set is a uniquely complete, dynamic 3D volume of the upper airway through a full respiratory cycle that can be processed for clinical and modeling analyses. A dynamic 3D CT technique of the upper airway is described that can be performed with a clinically reasonable radiation dose and sets a benchmark for future use.

A computer-aided system for automatic extraction of femur neck trabecular bone architecture using isotropic volume construction from clinical hip computed tomography images.

Hip fractures due to osteoporosis are increasing progressively across the globe. It is also difficult for those fractured patients to undergo dual-energy X-ray absorptiometry scans due to its complicated protocol and its associated cost. The utilisation of computed tomography for the fracture treatment has become common in the clinical practice. It would be helpful for orthopaedic clinicians, if they could get some additional information related to bone strength for better treatment planning. The aim of our study was to develop an automated system to segment the femoral neck region, extract the cortical and trabecular bone parameters, and assess the bone strength using an isotropic volume construction from clinical computed tomography images. The right hip computed tomography and right femur dual-energy X-ray absorptiometry measurements were taken from 50 south-Indian females aged 30-80 years. Each computed tomography image volume was re-constructed to form isotropic volumes. An automated system by incorporating active contour models was used to segment the neck region. A minimum distance boundary method was applied to isolate the cortical and trabecular bone components. The trabecular bone was enhanced and segmented using trabecular enrichment approach. The cortical and trabecular bone features were extracted and statistically compared with dual-energy X-ray absorptiometry measured femur neck bone mineral density. The extracted bone measures demonstrated a significant correlation with neck bone mineral density (r > 0.7, p < 0.001). The inclusion of cortical measures, along with the trabecular measures extracted after isotropic volume construction and trabecular enrichment approach procedures, resulted in better estimation of bone strength. The findings suggest that the proposed system using the clinical computed tomography images scanned with low dose could eventually be helpful in osteoporosis diagnosis and its treatment planning.

Histogram-Based Discrimination of Intravenous Contrast in Abdominopelvic Computed Tomography.

The aim of this study was to evaluate the accuracy of fully automated machine learning methods for detecting intravenous contrast in computed tomography (CT) studies of the abdomen and pelvis. A set of 591 labeled CT image volumes of the abdomen and pelvis was obtained from 5 different CT scanners, of which 434 (73%) were performed with intravenous contrast. A stratified split of this set was performed into training and test sets of 443 and 148 studies, respectively. Subsequently, support vector machine and logistic regression classifiers were trained using 5-fold cross-validation for parameter optimization. The best in-sample performance was seen with a support vector machine classifier with a χ kernel (98.9% accuracy); however, test set performance was similar across the trained classifiers, with 95% to 97% accuracy. Histogram-based automated classifiers for the presence of intravenous contrast are accurate and may be useful for verifying the accurate labeling of the presence of intravenous contrast in CT body studies.

One-Year Volume Stability of Human Facial Defects Filled With a β-Tricalcium Phosphate–Hydroxyl Apatite Mixture (Atlantik)

We investigated the applicability and 1-year stability of a β-tricalcium phosphate-hydroxyl apatite mixture (Atlantik) for secondary reconstruction of craniofacial defects and the application of OsiriX in evaluating bone and implant volumes. We included 6 patients (25-59 years) with craniofacial defects. A computed tomography scan was made preoperative, directly postoperative, and at least 1 year postoperative to evaluate volume changes. OsiriX was used to quantify volumes of the implanted Atlantik. Measurements were performed by 2 independent investigators and analyzed by calculating both Pearson correlation and interclass correlation coefficient. After 1 year, the mean volume reduction of the implanted Atlantik was 9.8%. The absolute volume reduction in 1 year was 0.38 cm (range, 0.10-0.69 cm(3)). Pearson correlation test was 0.996, with a significance level of P < 0.01, and the interclass correlation coefficient was 0.998. Atlantik is a stable osteoconductive material for the repair of various craniofacial defects. There is a reduction of only 10% of the augmented volume in the long term. Applying OsiriX for computed tomography image volume analysis proved to be a well-reproducible technique.

Development of an ultrasound phantom for spinal injections with 3-dimensional printing.

This report describes a method for producing anatomically detailed, low-cost ultrasound phantoms of the spine with 3-dimensional printing. An implementation that involves representing a portion of the lumbar spine and the ligamentum flavum with 2 different printing materials and the surrounding soft tissues with agar gel is presented. A computed tomography image volume of a patient with normal spinal anatomy was segmented to isolate the spine. Segments representing the ligamentum flavum and a supporting pedestal were digitally added, and the result was printed with a 3-dimensional printer. The printed spine was embedded in agar gel as a soft tissue component. Ultrasound images of the phantom were acquired and compared with those acquired from a human patient. The sonographic appearances of the phantom compared favorably with those observed from the human patient. The soft tissue component was suitable for needle insertions and could be remade replacing the agar. Ultrasound phantoms that are derived directly from patient anatomy have strong potential as learning tools for ultrasound-guided spinal insertions, and they could be used as preprocedural planning tools in cases involving pathologies, implants, or abnormal anatomies. Three-dimensional printing is a promising method for producing low-cost phantoms with designs that can be readily shared across clinical institutions.

Optimization of the scanning technique and diagnosis of pulmonary nodules with first-pass 64-detector-row perfusion VCT

ObjectiveThe aims of this study were to optimize the scanning technique of first-pass 64-detector-row perfusion volume computed tomography imaging, to evaluate the effectiveness and stability of this scan protocol, and lastly to evaluate the differential diagnosis ability of perfusion imaging in solitary pulmonary nodules (SPNs). MethodsA total of 144 patients with SPNs underwent perfusion scan with 64-slice spiral CT scanner. The CT perfusion imaging was analyzed for time–density curve, perfusion parametric maps, and the respective perfusion parameters. We then analyzed the main factors concerning the imaging quality and evaluated the effectiveness of scan protocol by determining the receiver operating characteristic (ROC) curve, diagnostic efficacy, and odds ratio as well as the stability of scan protocol by consistency analysis. Immunohistochemical findings of microvessel density measurement and vascular endothelial growth factor expression were evaluated. ResultsThe total sensitivity, specificity, accuracy, positive predictive value, negative predictive value, likelihood ratio, and the area under ROC curve during 5–45-s scan period were 78.95%, 82.4%, 80.6%, 83.3%, 77.8%, 4.620, 0.280, and 0.840, respectively, and Kappa value was 0.894. The diagnostic efficacy of CT pulmonary perfusion was significantly higher than during 0–40-s scan period. The parameter values in different nodules were different. ConclusionThe optimized 5–45-s scan period of CT pulmonary perfusion imaging is effective in pathologic diagnosis and has good stability, worthy of being popularized. Lung perfusion CT could be a promising and feasible method for differentiation of SPNs.

An Open Environment CT-US Fusion for Tissue Segmentation during Interventional Guidance

Therapeutic ultrasound (US) can be noninvasively focused to activate drugs, ablate tumors and deliver drugs beyond the blood brain barrier. However, well-controlled guidance of US therapy requires fusion with a navigational modality, such as magnetic resonance imaging (MRI) or X-ray computed tomography (CT). Here, we developed and validated tissue characterization using a fusion between US and CT. The performance of the CT/US fusion was quantified by the calibration error, target registration error and fiducial registration error. Met-1 tumors in the fat pads of 12 female FVB mice provided a model of developing breast cancer with which to evaluate CT-based tissue segmentation. Hounsfield units (HU) within the tumor and surrounding fat pad were quantified, validated with histology and segmented for parametric analysis (fat: −300 to 0 HU, protein-rich: 1 to 300 HU, and bone: HU>300). Our open source CT/US fusion system differentiated soft tissue, bone and fat with a spatial accuracy of ∼1 mm. Region of interest (ROI) analysis of the tumor and surrounding fat pad using a 1 mm2 ROI resulted in mean HU of 68±44 within the tumor and −97±52 within the fat pad adjacent to the tumor (p<0.005). The tumor area measured by CT and histology was correlated (r2 = 0.92), while the area designated as fat decreased with increasing tumor size (r2 = 0.51). Analysis of CT and histology images of the tumor and surrounding fat pad revealed an average percentage of fat of 65.3% vs. 75.2%, 36.5% vs. 48.4%, and 31.6% vs. 38.5% for tumors <75 mm3, 75–150 mm3 and >150 mm3, respectively. Further, CT mapped bone-soft tissue interfaces near the acoustic beam during real-time imaging. Combined CT/US is a feasible method for guiding interventions by tracking the acoustic focus within a pre-acquired CT image volume and characterizing tissues proximal to and surrounding the acoustic focus.

Comparison of two methods for quantitative assessment of mandibular asymmetry using cone beam computed tomography image volumes

The aim of this study was to compare two methods of measuring mandibular asymmetry. The first method uses mirroring of the mandible in the midsagittal plane; the second uses mirroring of the mandible and registration on the cranial base. Surface models were constructed from cone beam CT (CBCT) scans of 50 patients with asymmetry. For the first approach, a midsagittal plane was defined for each patient as the plane passing through nasion, anterior nasal spine and basion. Mirrors for both halves of the mandible were created. The second approach consisted of mirroring the image volume by flipping the left and right sides and then registering the mirrored image onto the cranial base using a mutual information maximization method. Surface distances between hemimandibles and mirrors were calculated for nine regions. There was no statistically significant difference between the mean surface distance measurements obtained with the two approaches and when comparing both halves in most areas. Both mirroring techniques provided similar quantification of mandibular asymmetry in this cohort.

Prospective evaluation of intraoperative computed tomography imaging for endoscopic sinonasal and skull‐base surgery

The objective of this study was to prospectively evaluate the clinical impact of intraoperative computed tomography (CT) imaging on endoscopic sinonasal and skull base procedures. A total of 49 patients were enrolled after informed consent from December 2009 to May 2010. Patients underwent intraoperative volume CT imaging (xCAT, Xoran Technologies, Ann Arbor, MI) at the conclusion of their proposed surgery. The mean age was 48.6 years with male:female ratio of 1.3:1. Surgical procedures included revision or primary endoscopic sinonasal surgery (ESS) (36), endoscopic benign or malignant tumor resection (10), endoscopic mucocele drainage (2), and endoscopic tumor biopsy (1). The mean Lund-Mackay (L-M) score was 10.6 (range 1-21). The indications for intraoperative imaging included extent of paranasal sinus dissection in 38 (77.6%), extent of tumor resection in 11 (22.4%), adequacy of mucocele drainage in 3 (6.1%), and frontal stent position in 2 (4.1%) cases. Average acquisition time was 5.3 minutes. The CT acquisition quality was deemed excellent in 24 (49.0%), good in 15 (30.6%), fair in 5 (10.2%), and unattainable in 5 (10.2%) cases. Additional interventions were performed in 8 of 44 cases (18%) based on the intraoperative CT dataset. Analysis of predictive factors for additional intervention, including presence of polyps, presence of tumor, previous surgery, use of image guidance, and CT quality did not reach statistical significance. Intraoperative CT scanning may hold important utility in selected endoscopic sinonasal and skull base procedures with additional interventions being performed in 18% of cases. The present study was unable to identify specific factors that would preoperatively predict the need for intraoperative imaging. Future clinical trials should include a multi-institutional design to better delineate these important variables.

An Initial Study of Three‐Dimensional Perfused Blood Volume Computed Tomography Imaging of Patients with Anterior Circulation Hyperacute Cerebral Infarction

To evaluate the value of three-dimensional (3D) whole brain perfused volume computed tomography (3D PBV CT) based on CT angiography (CTA) data in patients with hyperacute cerebral infarction. Thirty-five patients who had suffered stroke with anterior circulation within 3 hours underwent nonenhanced CT (NECT) scanning and CTA. Neuro PBV, a 3D software, was utilized to process the raw CTA data and a PBV image of the brain was obtained. Magnetic resonance imaging (MRI) was performed 6 hours after CT imaging. The volume and quantity of the ischemic lesion on 3D PBV and NECT were compared with MR-diffuse-weighted imaging (DWI). The numbers of cerebral infarcts detected by MRI, PBV, and NECT were 40, 38 and 16, respectively. The results of kappa analysis between NECT and PBV with MR were -.0128 and .7622, and a paired t-test analysis for the measurement of infarct volume between PBV and MRI was t = 7.249, P > .05. The lesions that were not detected by PBV volumes less than 4.5 cm(3). 3D PBV CT has the potential to assess the full extent of an ischemic stroke at an early stage, whereas PBV is limited to the detection of small infarcts. The 3D PBV CT technique based on CTA data requires no additional radiation exposure or contrast medium injection, and can be performed in a short period of time.

Comparison of two methods for quantitative assessment of mandibular asymmetry using cone beam computed tomography image volumes

Comparison of two methods for quantitative assessment of mandibular asymmetry using cone beam computed tomography image volumes

Comparing axial CT slices in quantized N-dimensional SURF descriptor space to estimate the visible body region

Abstract In this paper, a method is described to automatically estimate the visible body region of a computed tomography (CT) volume image. In order to quantify the body region, a body coordinate (BC) axis is used that runs in longitudinal direction. Its origin and unit length are patient-specific and depend on anatomical landmarks. The body region of a test volume is estimated by registering it only along the longitudinal axis to a set of reference CT volume images with known body coordinates. During these 1D registrations, an axial image slice of the test volume is compared to an axial slice of a reference volume by extracting a descriptor from both slices and measuring the similarity of the descriptors. A slice descriptor consists of histograms of visual words. Visual words are code words of a quantized feature space and can be thought of as classes of image patches with similar appearance. A slice descriptor is formed by sampling a slice on a regular 2D grid and extracting a Speeded Up Robust Features (SURF) descriptor at each sample point. The codebook, or visual vocabulary, is generated in a training step by clustering SURF descriptors. Each SURF descriptor extracted from a slice is classified into the closest visual word (or cluster center) and counted in a histogram. A slice is finally described by a spatial pyramid of such histograms. We introduce an extension of the SURF descriptors to an arbitrary number of dimensions ( N-SURF). Here, we make use of 2-SURF and 3-SURF descriptors. Cross-validation on 84 datasets shows the robustness of the results. The body portion can be estimated with an average error of 15.5 mm within 9 s. Possible applications of this method are automatic labeling of medical image databases and initialization of subsequent image analysis algorithms.

A Computer-Assisted System for Diagnostic Workstations: Automated Bone Labeling for CT Images

Although accurate information on thoracolumbar bone structure is essential when computed tomography (CT) images are examined, there is no automated method of labeling all the vertebrae and ribs on a CT scan. We are developing a computer-aided diagnosis system that labels ribs and thoracolumbar vertebrae automatically and have evaluated its accuracy. A candidate bone was extracted from the CT image volume data by pixel thresholding and connectivity analysis. All non-bony anatomical structures were removed using a linear discriminate of distribution of CT values and anatomical characteristics. The vertebrae were separated from the ribs on the basis of their distances from the centers of the vertebral bodies. Finally, the thoracic cage and lumbar vertebrae were extracted, and each vertebra was labeled with its own anatomical number by histogram analysis along the craniocaudal midline. The ribs were labeled in a similar manner, based on location data. Twenty-three cases were used for accuracy comparison between our method and the radiologist's. The automated labeling of the thoracolumbar vertebrae was concordant with the judgments of the radiologist in all cases, and all but the first and second ribs were labeled correctly. These two ribs were frequently misidentified, presumably because of pericostal anatomical clutter or high densities of contrast material in the injected veins. We are confident that this system can contribute usefully as part of a picture archiving and communication system workstation, though further technical improvement is required for identification of the upper ribs.