Semi-automatic Segmentation Algorithm Research Articles

Statistical validation of image segmentation quality based on a spatial overlap index 1: scientific reports

Rationale and Objectives. To examine a statistical validation method based on the spatial overlap between two sets of segmentations of the same anatomy. Materials and Methods. The Dice similarity coefficient (DSC) was used as a statistical validation metric to evaluate the performance of both the reproducibility of manual segmentations and the spatial overlap accuracy of automated probabilistic fractional segmentation of MR images, illustrated on two clinical examples. Example 1: 10 consecutive cases of prostate brachytherapy patients underwent both preoperative 1.5T and intraoperative 0.5T MR imaging. For each case, 5 repeated manual segmentations of the prostate peripheral zone were performed separately on preoperative and on intraoperative images. Example 2: A semi-automated probabilistic fractional segmentation algorithm was applied to MR imaging of 9 cases with 3 types of brain tumors. DSC values were computed and logit-transformed values were compared in the mean with the analysis of variance (ANOVA). Results. Example 1: The mean DSCs of 0.883 (range, 0.876–0.893) with 1.5T preoperative MRI and 0.838 (range, 0.819–0.852) with 0.5T intraoperative MRI ( P <.001) were within and at the margin of the range of good reproducibility, respectively. Example 2: Wide ranges of DSC were observed in brain tumor segmentations: Meningiomas (0.519–0.893), astrocytomas (0.487–0.972), and other mixed gliomas (0.490–0.899). Conclusion. The DSC value is a simple and useful summary measure of spatial overlap, which can be applied to studies of reproducibility and accuracy in image segmentation. We observed generally satisfactory but variable validation results in two clinical applications. This metric may be adapted for similar validation tasks.

Statistical validation based on parametric receiver operating characteristic analysis of continuous classification data 1

Rationale and Objectives. The accuracy of diagnostic test and imaging segmentation is important in clinical practice because it has a direct impact on therapeutic planning. Statistical validations of classification accuracy was conducted based on parametric receiver operating characteristic analysis, illustrated on three radiologic examples. Materials and Methods. Two parametric models were developed for diagnostic or imaging data. Example 1: A semi-automated fractional segmentation algorithm was applied to magnetic resonance imaging of nine cases of brain tumors. The tumor and background pixel data were assumed to have bi-beta distributions. Fractional segmentation was validated against an estimated composite pixel-wise gold standard based on multi-reader manual segmentations. Example 2: The predictive value of 100 cases of spiral computed tomography of ureteral stone sizes, distributed as bi-normal after a nonlinear transformation, under two treatment options received. Example 3: One hundred eighty cases had prostate-specific antigen levels measured in a prospective clinical trial. Radical prostatectomy was performed in all to provide a binary gold standard of local and advanced cancer stages. Prostate-specific antigen level was transformed and modeled by bi-normal distributions. In all examples, areas under the receiver operating characteristic curves were computed. Results. The areas under the receiver operating characteristic curves were: Example 1: Fractional segmentation of magnetic resonance imaging of brain tumors: meningiomas (0.924–0.984); astrocytomas (0.786–0.986); and other low-grade gliomas (0.896–0.983). Example 3: Ureteral stone size for treatment planning (0.813). Example 2: Prostate-specific antigen for staging prostate cancer (0.768). Conclusion. All clinical examples yielded fair to excellent accuracy. The validation metric area under the receiver operating characteristic curves may be generalized to evaluating the performances of several continuous classifiers related to imaging.

Computed SAR and thermal elevation in a 0.25-mm 2-D model of the human eye and head in response to an implanted retinal stimulator. I. Models and methods

Retinitis pigmentosa and age-related macular degeneration lead to blindness through progressive loss of retinal photoreceptors. Attempts are under way to construct a visual prosthesis to recover a limited sense of vision for these patients with the aid of implantable electronic devices. The function of these microchips is to provide electrical stimulation to existing viable retinal tissues - living ganglion and bipolar cells - using an array of on-chip stimulus circuits, while the dominant mechanism for power and data communication for these implanted devices has been wireless inductive telemetry using coils. This paper describes methods and models used to estimate the heating induced in the human eye and surrounding head tissues subject to the operation of this retinal prosthesis. A two-dimensional 0.25-mm high-resolution human head model has been developed with the aid of a new semiautomatic graphical segmentation algorithm. Finite-difference-based numerical methods for both electromagnetic and thermal modeling have been used to determine the influence of the specific absorption rate (associated with 2-MHz inductive coupling to the implant) and of stimulator integrated circuit (IC) power on tissue heating under different operational conditions and different hypothesis on choroidal blood flow and properties of the complex implanted circuitry. Results, provided in Part II of this paper, show that temperature increases of approximately 0.6 and 0.4/spl deg/C are induced in the midvitreous of the human eye in the absence and presence of choroidal blood flow, respectively, for a 60-electrode retinal prosthesis chip. Correspondent temperature rises of approximately 0.19 and 0.004/spl deg/C on the retina are obtained for these cases. Comparison with in vivo experimental measurements on intraocular heating in dog eyes shows good agreement.

Registration of 3D CT angiography and cardiac MR images in coronary artery disease patients.

A method for the registration of 3D cardiac CT angiography (CTA) and magnetic resonance (MR) data sets based on their myocardial epicardial surfaces is introduced. The approach relies on temporally registered data sets obtained based on the electrocardiogram recorded during the CTA acquisition and the timing characteristics of the MR acquisition. The myocardial epicardial surfaces were identified in the reformatted CTA and MR data sets using a 3D semi-automated segmentation algorithm. This algorithm was implemented, evaluated on clinical data, and compared to a set of manual outlines during the course of this study. The registration of the CTA and MR data sets was based on the iterative closest point algorithm, which minimizes the distance between the surfaces defined by the epicardial outlines in each data set. The proposed technique was applied to data obtained from 11 patients with coronary artery disease. The CTA data was reformatted based on the registration results and the location of the MR imaging planes. The resulting CTA-MR image pairs were evaluated qualitatively by two experts, who graded the majority of the cases as either excellent or acceptable (11 of 11 cases for one reader, and 9 of 11 for the other). The results were evaluated quantitatively based on the distance between the registered epicardial surfaces. The quantitative measures indicated that the registered surfaces were within two pixels of one another (on average). The registration results were used to generate combined 3D renderings of information extracted from both data sets for visualization purposes.

ROPES: a semiautomated segmentation method for accelerated analysis of three-dimensional echocardiographic data.

Echocardiography (cardiac ultrasound) is today the predominant technique for quantitative assessment of cardiac function and valvular heart lesions. Segmentation of cardiac structures is required to determine many important diagnostic parameters. As the heart is a moving organ, reliable information can be obtained only from three-dimensional (3-D) data over time (3-D + time = 4-D). Due to their size, the resulting four-dimensional (4-D) data sets are not reasonably accessible to simple manual segmentation methods. Automatic segmentation often yields unsatisfactory results in a clinical environment, especially for ultrasonic images. We describe a semiautomated segmentation algorithm (ROPES) that is able to greatly reduce the time necessary for user interaction and its application to extract various parameters from 4-D echocardiographic data. After searching for candidate contour points, which have to fulfill a multiscale edge criterion, the candidates are connected by minimizing a cost function to line segments that then are connected to form a closed contour. The contour is automatically checked for plausibility. If necessary, two correction methods that can also be used interactively are applied (fitting of other line segments into the contour and searching for additional candidates with a relaxed criterion). The method is validated using in vivo transesophageal echocardiographic data sets.

High-resolution magnetic resonance imaging tracks changes in organ and tissue mass in obese and aging rats

Magnetic resonance imaging (MRI) has the ability to discriminate between various soft tissues in vivo. Whole body, specific organ, total adipose tissue (TAT), intra-abdominal adipose tissue (IAAT), and skeletal muscle (SM) weights determined by MRI were compared with weights determined by dissection and chemical analysis in two studies with male Sprague-Dawley rats. A 4.2-T MRI machine acquired high-resolution, in vivo, longitudinal whole body images of rats as they developed obesity or aged. Weights of the whole body and specific tissues were determined using computer image analysis software, including semiautomatic segmentation algorithms for volume calculations. High correlations were found for body weight (r = 0.98), TAT (r = 0.99), and IAAT (r = 0.98) between MRI and dissection and chemical analyses. MRI estimated the weight of the brain, kidneys, and spleen with high accuracy (r > 0.9), but overestimated IAAT, SM, and liver volumes. No differences were detected in organ weights using MRI and dissection measurements. Longitudinal MRI measurements made during the development of obesity and aging accurately represented changes in organ and tissue mass.

The segmented regional volumes of the cerebrum and cerebellum in boys with Tourette syndrome.

Neuropathological deficits are an etiological factor in Tourette syndrome (TS), and implicate a network linking the basal ganglia and the cerebrum, not a particular single brain region. In this study, the volumes of 20 cerebral and cerebellar regions and their symmetries were measured in normal boys and TS boys by brain magnetic resonance imaging. Brain magnetic resonance images were obtained prospectively in 19 boys with TS and 17 age-matched normal control boys. Cerebral and cerebellar regions were segmented to gray and white fractions using algorithm for semi-automated fuzzy tissue segmentation. The frontal, parietal, temporal, and the occipital lobes and the cerebellum were defined using the semiautomated Talairach atlas-based parcellation method. Boys with TS had smaller total brain volumes than control subjects. In the gray matter, although the smaller brain volume was taken into account, TS boys had a smaller right frontal lobe and a larger left frontal lobe and increased normal asymmetry (left>right). In addition, TS boys had more frontal lobe white matter. There were no significant differences in regions of interest of the parietal, temporal, or the occipital lobes or the cerebellum. These findings suggest that boys with TS may have neuropathological abnormalities in the gray and the white matter of the frontal lobe.

Cerebrospinal fluid dynamics and relation with blood flow: a magnetic resonance study with semiautomated cerebrospinal fluid segmentation.

To investigate and measure temporal and amplitude aspects of blood and cerebrospinal fluid (CSF) flow waveform relations. A cine phase-contrast magnetic resonance imaging pulse sequence was used to measure blood and CSF flow in 16 healthy subjects aged 27 +/- 4 years. A semiautomated segmentation algorithm was developed to study CSF flow. Standard deviations of the aqueductal and cervical flow measurements carried out by five observers were 1% and 4%, respectively. The peak systolic arterial flow was 1087 +/- 169 mL/min, and the peak cervical CSF flush (173 +/- 59 mL/min) occurred at 5% +/- 3% of the cardiac cycle after the internal carotid systolic peak flow. Peak aqueductal flush flow (13 +/- 5 mL/min) occurred at 21% +/- 7% of the cardiac cycle after the internal carotid systolic peak flow. The CSF segmentation algorithm is reproducible. Brain expansion was quickly regulated by a major extracerebral CSF flush flow, whereas ventricular CSF made only a very small contribution.

Semi-automated color segmentation of anatomical tissue

We propose a semi-automated region-based color segmentation algorithm to extract anatomical structures, including soft tissues, in the color anatomy slices of the Visible Human data. Our approach is based on repeatedly dividing an image into regions using Voronoi diagrams and classifying the regions based on experimental classification statistics. The user has the option of reclassifying regions in order to improve the final boundary. Our results indicate that the algorithm can find accurate outlines in a small number of iterations and that manual interaction can markedly improve the outline. This approach can be extended to 3D color segmentation.

Precision MRI-based joint surface and cartilage density analysis of the knee joint using rapid water-excitation sequence and semi-automatic segmentation algorithm

The aim of this study was to analyse the precision of three-dimensional joint surface and cartilage thickness measurements in the knee, using a fast, high-resolution water-excitation sequence and a semiautomated segmentation algorithm. The knee joint of 8 healthy volunteers, aged 22 to 29 years, were examined at a resolution of 1.5 mm x 0.31 mm x 0.31 mm, with four sagittal data sets being acquired after repositioning the joint. After semiautomated segmentation with a B-spline Snake algorithm and 3D reconstruction of the patellar, femoral and tibial cartilages, the joint surface areas (triangulation), cartilage volume, and mean and maximum thickness (Euclidean distance transformation) were analysed, independently of the orientation of the sections. The precision (CV%) for the surface areas was 2.1 to 6.6%. The mean cartilage thickness and cartilage volume showed coefficients of 1.9 to 3.5% (except for the femoral condyles), the value for the medial femoral condyle being 9.1%, and for the lateral condyle 6.5%. For maximum thickness, coefficients of between 2.6 and 5.9% were found. In the present study we investigate for the first time the precision of MRI-based joint surface area measurements in the knee, and of cartilage thickness analyses in the femur. Using a selective water-excitation sequence, the acquisition time can be reduced by more than 50%. The poorer precision in the femoral condyles can be attributed to partial volume effects that occur at the edges of the joint surfaces with a sagittal image protocol. Since MRI is non-invasive, it is highly suitable for examination of healthy subjects (generation of individual finite element models, analysis of functional adaptation to mechanical stimulation, measurement of cartilage deformation in vivo) and as a diagnostic tool for follow-up, indication for therapy, and objective evaluation of new therapeutic agents in osteoarthritis.

Accelerating ‘Intelligent Scissors’ Using Slimmed Graphs

Abstract In this paper, we describe an acceleration technique for the semiautomatic image segmentation algorithm, “int elligent scissors”. Using intelligent scissors, the user can accurately and interactively extract an object from a digitized image. However, the original algorithm suffers from slow performance when large images are treated. In practice, pixels within the non-edge regions are seldom involved in the determination of boundaries (segmentation curves). If these pixels are removed before boundary determination, the performance of intelligent scissors can be improved. We generate a slimmed graph to achieve this goal with significant improvement in response time.

Optimization of tissue segmentation of brain MR images based on multispectral 3D feature maps

The purpose of this work was to optimize and increase the accuracy of tissue segmentation of the brain magnetic resonance (MR) images based on multispectral 3D feature maps. We used three sets of MR images as input to the in-house developed semi-automated 3D tissue segmentation algorithm: proton density (PD) and T 2-weighted fast spin echo and, T 1-weighted spin echo. First, to eliminate the random noise, non-linear anisotropic diffusion type filtering was applied to all the images. Second, to reduce the nonuniformity of the images, we devised and applied a correction algorithm based on uniform phantoms. Following these steps, the qualified observer “seeded” (identified training points) the tissue of interest. To reduce the operator dependent errors, cluster optimization was also used; this clustering algorithm identifies the densest clusters pertaining to the tissues. Finally, the images were segmented using k-NN (k-Nearest Neighborhood) algorithm and a stack of color-coded segmented images were created along with the connectivity algorithm to generate the entire surface of the brain. The application of pre-processing optimization steps substantially improved the 3D tissue segmentation methodology.

Usability of Semiautomatic Segmentation Algorithms for Tumor Volume Determination

Tumor volume is an important parameter for clinical decision making. At present, semiautomatic image segmentation is not a standard for tumor volumetry. The aim of this work was to investigate the usability of semiautomatic algorithms for tumor volume determination. Semiautomatic region- and volume-growing, isocontour, snakes, hierarchical, and histogram-based segmentation algorithms were tested for accuracy, contour variability, and time performance. The test were performed on a newly developed organic phantom for the simulation of a human liver and liver metastases. The real tumor volumes were measured by water displacement. These measured volumes were used as the gold standard for determining the accuracy of the algorithms. Variability of the segmented volumes ranging from 3.9 +/- 3.2% (isocontour algorithm) to 11.5 +/- 13.9% (hierarchical segmentation) was observed. The segmentation time per slice varied between 32 (volume-growing) and 72 seconds (snakes) on an IBM/RS6000 workstation. Only the region-growing and isocontour algorithms have the potential to be used for tumor volumetry. However, further improvements of these algorithms are necessary before they can be placed into clinical use.

Evidence For Genetic Variance in White Matter Hyperintensity Volume in Normal Elderly Male Twins

White matter hyperintensities (WMHs), as detected by MRI, are common among the elderly and are frequently interpreted as representing a subclinical form of ischemic brain damage. We used volumetric MR techniques to investigate the contribution of genes and the environment to measures of brain morphology in a sample of community dwelling elderly male twins. Brain MR (1.5 T) scans were obtained from 74 monozygotic (MZ) and 71 dizygotic (DZ), white, male, World War II veteran twins born in the United States and age 68 to 79 when scanned. MR quantification used a previously published semiautomated segmentation algorithm to segment brain images into total brain, cerebrospinal fluid (CSF), and WMH volumes. Twin pair covariances were computed for each measure, and structural equation genetic models were fitted to these data. Total cranial, brain parenchyma, CSF, and WMH volumes were highly correlated in MZ pairs, and correlations in MZ pairs were significantly greater than those in DZ pairs. Structural equation modeling indicated heritabilities of 91%, 92%, and 73%, respectively, for total cranial, brain parenchyma, and WMH volumes. Correction for age and head size reduced the heritability of brain parenchyma to 62% (95% confidence interval, 56% to 68%) and the heritability of WMH volume to 71% (95% confidence interval, 66% to 76%). Proband concordance rates for large amounts of WMH were 61% in MZ pairs and 38% in DZ pairs, compared with a prevalence of 15% in the entire sample. This study is the first to quantify the relative contribution of genetic and individual environmental influences to measures of brain morphology in the elderly.

Three-dimensional visualization of the mandible: A new method for presenting the periodontal status and diseases

A three-dimensional representation of an ideal human jaw was reconstructed from a series of 178 digitized photographic cross-sections. The two-dimensional images were taken from the slices of an artificial skull and have been digitized by means of a high resolution scanner with a spatial resolution of 860 dots per inch. On the basis of the sequence of cross-sections, a semi-automatic segmentation algorithm was developed to reduce the information to a quadriliteral surface-representation of the teeth and the bone. An algorithm was developed which simulates the individual pathology of a patient both on the basis of the findings stored in this patient's dental record and by using the representation of the reference jaw. The results of the modeling module were automatically prepared for rendering with visualization tools compatible to the RenderMan standard. This new method of presenting periodontal situations is especially helpful for the diagnostic support of periodontologists and for dental educational purposes.

MRI volumetrics in late life depression and alzheimer's disease

In order to examine the structural basis of late-life depression, we utilized MRIs to study cerebral hemispheric and CSF volumes (sulcal, ventricular and total) in 18 patients with late-life depression (TM, IIW), mean age 73.6 (SD 8.2), 34 subjects (17M, 17W), mean age 66 (SD 8.6) who met criteria for probable dementia of the Alzheimer type (DAT) and 29 healthy age-matched control (12M, 17W), mean age 67 (SD 8.5). All depressed subjects met DSM-III-R criteria for major depression, had Hamiiton Rating Scores for Depression >!5 and were cognitively intact at the time of the study. All subjects were medically stable and free of significant other neumpsychiatric illnesses. Proton density and 'r2 weighted MRI's were used to maximize the contrast between neural tissue and CSF. A semi-automated boundary program and segmentation algorithm was then used to differentiate brain tissue from CSF. On normalized measures of CSF volumes (venMcular, sulcal and total) in both hemispheres, depressed subjects had significantly larger volumes when compared to conIrol (p<0.0$), as did DAT patients. When depressed and DAT patients were compared, the right CSF (larger in those with DAT) was the only area where there was a statistically significant size difference. These preliminary data suggest that subjects with late-life depression have structural brain changes as determined by MRI. CSF volumetric measures in the depressed group were comparable in magnitude to those with DAT and significantly different from normal controls.