Low-contrast Objects Research Articles

TU-E-BRA-08: A Comprehensive Study on the Relationship between Image Quality and Imaging Dose in Low-Dose Cone Beam CT.

To investigate image quality as a function of number of projections and tube load per projection in compressive sensing (CS) based low-dose cone beam CT (CBCT), and achieve optimal low-dose scan protocols in image guided radiation therapy (IGRT). We have performed CS-based CBCT reconstruction with different combinations of number of projections (from 46 to 364) and mAs (from 0.2 to 2.4 mAs/view), which covers the whole clinically relevant range. Image quality is assessed in each case. On this basis, optimal scan protocols are analyzed according to various IGRT applications. Image quality degrades ∼10% when the imaging dose decreases from 400 to 100 total mAs, further ∼10% from 100 to 40 total mAs, and another ∼80% below 40 total mAs. Image quality on iso-low-dose lines at 36.8, 72.6, 109.2 and 145.6 total mAs varies 17.16%, 13.69%, 11.99% and 5.74% in terms RMSE with various scanning protocols. 1) In CS-based CBCT, image quality has little degradation with imaging dose> 100 total mAs. Optimal low-dose scan protocols likely fall in the range of 40-100 total mAs. 2) At a constant low-dose level, the scan protocol that with super sparse views (projection number < 50) is the most challenging case. 3) The optimal scan protocol is the combination of a medium number of projections and a medium level of mAs/view. This is more evident when the dose is ∼72.8 total mAs or below, and when the ROI is a low-contrast or high-resolution object. 4) The clinically acceptable lowest imaging dose level is task dependent. In our study, 72.8mAs is a safe dose level for visualizing low-contrast objects, while 12.2 total mAs is sufficient for detecting high-contrast objects of diameter greater than 3 mm. This work is supported in part by NIH (1R01CA154747-01), Varian Medical Systems through a Master Research Agreement, and the Thrasher Research Fund.

SU-E-I-56: Threshold Effect of ASIR Before Which Image Improve and After Which Image Degrades.

This study showed to what extent ASIR improves CT-image and to what extent it degrades it. In our study we used GE HD750 CT-scanner, Siemens Sensation CT-scanner, Catphan, PTW-pin-ion- chamber, CTDI-phantom. We measured the CT-dose using the PTW-pinion-chamber and CTDI-phantom. Image-quality and noise were evaluated using catphan and GE water phantom. Image noise reduce as higher levels of ASIR are applied. A phantom scan showed that 50%ASIR with 50% lower-dose (10.8mGy) achieved the same image noise of standard FBP image with full dose 21.7mGy (noise∼5). To confirm that the two same-noise images retain same image-quality, two scans were compared; one with full dose 260mAs(21.7mGy) and the other one with 50% lower dose 130mAs(10.8mGy). The results showed that ASIR failed to retain the same quality. For high contrast resolution, 50%ASIR reduced the resolution of patterns = 71p/cm, however it improved the detectability of patterns = 61p/cm. ASIR has degraded the CNR of the low-contrast-objects of = 5HU (CNR of 1.4 at 260mAs STND to CNR of 1.08 at 130mAs ASIR), however it improved the CNR of the low-contrast-objects of = 10HU (CNR of 2.35 at 260mAs STND to CNR of 2.63 at 130mAs ASIR). ASIR degraded the edges and killed some of the small objects. This shows that ASIR has a critical point of improve/degrade. Also, ASIR can improve images for the same dose, but with high levels of ASIR (e.g. 100%ASIR), cause disapear of small low contrast objects (e.g. 2mm). People think that ASIR only improves image and reduces patient dose. Our study showed that ASIR has some drawbacks. There is a threshold before wich ASIR is positive and after which ASIR is negative. Recently only GE provide ASIR in the market but our study showed that other CTs such as Siemens can do similar performance like ASIR.

Verification of DICOM GSDF in Complex Backgrounds

While previous research has determined the contrast detection threshold in medical images, it has focused on uniform backgrounds, has not used calibrated monitors, or has involved a low number of readers. With complex clinical images, how the Grayscale Standard Display Function (GSDF) affects the detection threshold and whether the median background intensity shift has been minimized by GSDF remains unknown. We set out to determine if the median background affected the detection of a low-contrast object in a clustered lumpy background, which simulated a mammography image, and to define the contrast detection threshold for these complex images. Clustered lumpy background images were created of different median intensities and disks of varying contrasts were inserted. A reader study was performed with 17 readers of varying skill level who scored with a five-point confidence scale whether a disk was present. The results were analyzed using reader operating characteristic (ROC) methodology. Contingency tables were used to determine the contrast detection threshold. No statistically significant difference was seen in the area under the ROC curve across all of the backgrounds. Contrast detection fell below 50% between +3 and +2gray levels. Our work supports the conclusion that Digital Imaging and Communications in Medicine GSDF calibrated monitors do perceptually linearize detection performance across shifts in median background intensity. The contrast detection threshold was determined to be +3gray levels above the background for an object of 1° visual angle.

Specific features of radio emissions of coronal holes based on eclipse and noneclipse observations at a solar activity minimum

Solar coronal holes (CHs) at a minimum of the 23rd activity cycle were investigated using Solar and Heliospheric Observatory (SOHO) data, ground-based observational data from the radio telescopes of the Kislovodsk Solar Station, Pulkovo Observatory (KSS PO), Russian Academy of Sciences, and radioheliograph data from the Nobeyama Observatory (Japan). The 2006–2008 period was characterized by a small number of active regions on the solar disk; nevertheless, this period is favorable for studying low-contrast objects in the radio band (CHs). We investigate the evolution of CH areas, the location of CHs on the solar disk, and the features of their radiation in the radio band. We present the results of observations of the total (March 29, 2006) and partial (August 1, 2008) solar eclipses by the RT-3 and RT-2 radio telescopes of the KSS PO. Based on the eclipse observation data, compact sources were identified on the solar disk and the contribution of CHs to the integral radio emission flux was estimated. A rare effect (increased radio emissions of high-latitude CHs at a wavelength of 4.9 cm) was revealed, which may be caused by X-ray bright points in CHs. Here, polar CHs are characterized by low levels of radio emission.

A comprehensive study on the relationship between the image quality and imaging dose in low-dose cone beam CT

While compressed sensing (CS)-based algorithms have been developed for the low-dose cone beam CT (CBCT) reconstruction, a clear understanding of the relationship between the image quality and imaging dose at low-dose levels is needed. In this paper, we qualitatively investigate this subject in a comprehensive manner with extensive experimental and simulation studies. The basic idea is to plot both the image quality and imaging dose together as functions of the number of projections and mAs per projection over the whole clinically relevant range. On this basis, a clear understanding of the tradeoff between the image quality and imaging dose can be achieved and optimal low-dose CBCT scan protocols can be developed to maximize the dose reduction while minimizing the image quality loss for various imaging tasks in image-guided radiation therapy (IGRT). Main findings of this work include (1) under the CS-based reconstruction framework, image quality has little degradation over a large range of dose variation. Image quality degradation becomes evident when the imaging dose (approximated with the x-ray tube load) is decreased below 100 total mAs. An imaging dose lower than 40 total mAs leads to a dramatic image degradation, and thus should be used cautiously. Optimal low-dose CBCT scan protocols likely fall in the dose range of 40–100 total mAs, depending on the specific IGRT applications. (2) Among different scan protocols at a constant low-dose level, the super sparse-view reconstruction with the projection number less than 50 is the most challenging case, even with strong regularization. Better image quality can be acquired with low mAs protocols. (3) The optimal scan protocol is the combination of a medium number of projections and a medium level of mAs/view. This is more evident when the dose is around 72.8 total mAs or below and when the ROI is a low-contrast or high-resolution object. Based on our results, the optimal number of projections is around 90 to 120. (4) The clinically acceptable lowest imaging dose level is task dependent. In our study, 72.8 mAs is a safe dose level for visualizing low-contrast objects, while 12.2 total mAs is sufficient for detecting high-contrast objects of diameter greater than 3 mm.

An Adaptive Gaussian Mixture Model for Non-rigid Image Registration

This paper presents a new model based on statistical and variational methods for non-rigid image registration. It can be viewed as an improvement of the intensity-based model whose dissimilarity term is based on minimization of the so-called sum of squared difference(SSD). In the proposed model, it is assumed that the residue of two images can be described as a mixture of Gaussian distributions. Then we incorporate the features of variational regularization methods and expectation-maximization(EM) algorithm, and propose the new model. The novelty is the introduction of two weighting functions and some control parameters in dissimilarity term. The weighting functions could identify low and high contrast objects of the residue automatically and effectively, and the control parameters help to improve the robustness of the model to the choice of regularization parameters. By the introduced parameters and weighting functions, the algorithm could locally adjust the behavior of deformation in different contrast regions. Numerical experimental results of 2D synthetic and 3D MR brain images demonstrate the efficiency and accuracy of the proposed approach compared with other methods.

Dual-mode detection for foreground segmentation in low-contrast video images

In video surveillance, the detection of foreground objects in an image sequence from a still camera is critical for object tracking, activity recognition, and behavior understanding. The widely used background updating models such as single Gaussian and mixture of Gaussians are based mainly on the mean gray level of a given observation period, which could be inevitably affected by outliers and noise in the images. The mean from the average of background pixel values and foreground pixel values of consecutive image frames in an observed period will not represent the true background in the scene. Thus, the mean background models cannot detect low-contrast objects and promptly respond to sudden light changes. In this paper, a dual-mode scheme for foreground segmentation is proposed. The mode is based on the most frequently occurring gray level of observed consecutive image frames, and is used to represent the background in the scene. In order to accommodate the dynamic changes of a background, the proposed method uses a dual-mode model for background representation. The dual-mode model can represent two main states of the background and detect a more complete silhouette of the foreground object in the dynamic background. The proposed method can promptly calculate the exact gray level mode of individual pixels in image sequences by simply dropping the last image frame and adding the current image in an observed period. The comparative evaluation of foreground segmentation methods is performed on the Microsoft's Wallflower dataset. The results show that the proposed method can quickly respond to illumination changes and well extract foreground objects in a low-contrast background.

Effect of Low Tube Voltage on Image Quality, Radiation Dose, and Low-Contrast Detectability at Abdominal Multidetector CT: Phantom Study

Purpose. To investigate the effect of low tube voltage (80 kV) on image quality, radiation dose, and low-contrast detectability (LCD) at abdominal computed tomography (CT). Materials and Methods. A phantom containing low-contrast objects was scanned with a CT scanner at 80 and 120 kV, with tube current-time product settings at 150–650 mAs. The differences between image noise, contrast-to-noise ratio (CNR), and scores of LCD obtained with 80 kV at 150–650 mAs and those obtained with 120 kV at 300 mAs were compared respectively. Results. The image noise substantially increased with low tube voltage. However, with identical dose, use of 80 kV resulted in higher CNR compared with CNR at 120 kV. There were no statistically significant difference in CNR and scores of LCD between 120 kV at 300 mAs and 80 kV at 550–650 mAs (P > 0.05). The relative dose delivered at 80 kV ranged from 58% at 550 mAs to 68% at 650 mAs. Conclusion. With a reduction of the tube voltage from 120 kV to 80 kV at abdominal CT, the radiation dose can be reduced by 32% to 42% without degradation of CNR and LCD.

Visual Backward Masking Performance in Young Adult Emmetropes and Myopes

To investigate how temporal processing is altered in myopia and during myopic progression. In backward visual masking, a target's visibility is reduced by a mask presented quickly after the target. Thirty emmetropes, 40 low myopes, and 22 high myopes aged 18 to 26 years completed location and resolution masking tasks. The location task examined the ability to detect letters with low contrast and large stimulus size. The resolution task involved identifying a small letter and tested resolution and color discrimination. Target and mask stimuli were presented at nine short interstimulus intervals (12 to 259 ms) and at 1000 ms (long interstimulus interval condition). In comparison with emmetropes, myopes had reduced ability in both locating and identifying briefly presented stimuli but were more affected by backward masking for a low contrast location task than for a resolution task. Performances of low and high myopes, as well as stable and progressing myopes, were similar for both masking tasks. Task performance was not correlated with myopia magnitude. Myopes were more affected than emmetropes by masking stimuli for the location task. This was not affected by magnitude or progression rate of myopia, suggesting that myopes have the propensity for poor performance in locating briefly presented low contrast objects at an early stage of myopia development.

Image Quality Evaluation of New Image Reconstruction Methods Applying the Iterative Reconstruction

The purpose of this study was to evaluate the image quality of an iterative reconstruction method, the iterative reconstruction in image space (IRIS), which was implemented in a 128-slices multi-detector computed tomography system (MDCT), Siemens Somatom Definition Flash (Definition). We evaluated image noise by standard deviation (SD) as many researchers did before, and in addition, we measured modulation transfer function (MTF), noise power spectrum (NPS), and perceptual low-contrast detectability using a water phantom including a low-contrast object with a 10 Hounsfield unit (HU) contrast, to evaluate whether the noise reduction of IRIS was effective. The SD and NPS were measured from the images of a water phantom. The MTF was measured from images of a thin metal wire and a bar pattern phantom with the bar contrast of 125 HU. The NPS of IRIS was lower than that of filtered back projection (FBP) at middle and high frequency regions. The SD values were reduced by 21%. The MTF of IRIS and FBP measured by the wire phantom coincided precisely. However, for the bar pattern phantom, the MTF values of IRIS at 0.625 and 0.833 cycle/mm were lower than those of FBP. Despite the reduction of the SD and the NPS, the low-contrast detectability study indicated no significant difference between IRIS and FBP. From these results, it was demonstrated that IRIS had the noise reduction performance with exact preservation for high contrast resolution and slight degradation of middle contrast resolution, and could slightly improve the low contrast detectability but with no significance.

Intelligent Recognition of Lung Nodule Combining Rule-based and C-SVM Classifiers

Abstract Computer-aided detection(CAD) system for lung nodules plays the important role in the diagnosis of lung cancer. In this paper, an improved intelligent recognition method of lung nodule in HRCT combing rule-based and cost-sensitive support vector machine(C-SVM) classifiers is proposed for detecting both solid nodules and ground-glass opacity(GGO) nodules(part solid and nonsolid). This method consists of several steps. Firstly, segmentation of regions of interest(ROIs), including pulmonary parenchyma and lung nodule candidates, is a difficult task. On one side, the presence of noise lowers the visibility of low-contrast objects. On the other side, different types of nodules, including small nodules, nodules connecting to vasculature or other structures, part-solid or nonsolid nodules, are complex, noisy, weak edge or difficult to define the boundary. In order to overcome the difficulties of obvious boundary-leak and slow evolvement speed problem in segmentatioin of weak edge, an overall segmentatio...

Nonlocal means-based denoising for medical images.

Medical images often consist of low-contrast objects corrupted by random noise arising in the image acquisition process. Thus, image denoising is one of the fundamental tasks required by medical imaging analysis. Nonlocal means (NL-means) method provides a powerful framework for denoising. In this work, we investigate an adaptive denoising scheme based on the patch NL-means algorithm for medical imaging denoising. In contrast with the traditional NL-means algorithm, the proposed adaptive NL-means denoising scheme has three unique features. First, we use a restricted local neighbourhood where the true intensity for each noisy pixel is estimated from a set of selected neighbouring pixels to perform the denoising process. Second, the weights used are calculated thanks to the similarity between the patch to denoise and the other patches candidates. Finally, we apply the steering kernel to preserve the details of the images. The proposed method has been compared with similar state-of-art methods over synthetic and real clinical medical images showing an improved performance in all cases analyzed.

Flatfield Correction Optimization for Energy Selective X-Ray Imaging With Medipix3

Pixelated photon counting semiconductor X-ray detectors like the Medipix feature adjustable energy thresholds allowing selective counting of photons of a specified energy. This development permits for energy selective X-ray imaging with advanced material information. Furthermore the photon counting function principle of these detectors allows X-ray imaging with reduced noise, providing contrast improvement in low contrast objects. The aim of this study is to analyze the behavior of the new Medipix3 detector, especially regarding flatfield correction for X-ray imaging applications. First high resolution low contrast X-ray images and energy selective X-ray images acquired with the Medipix3 detector are presented as well in this paper.

An Approach to Improve the Quality of Infrared Images of Vein-Patterns

This study develops an approach to improve the quality of infrared (IR) images of vein-patterns, which usually have noise, low contrast, low brightness and small objects of interest, thus requiring preprocessing to improve their quality. The main characteristics of the proposed approach are that no prior knowledge about the IR image is necessary and no parameters must be preset. Two main goals are sought: impulse noise reduction and adaptive contrast enhancement technologies. In our study, a fast median-based filter (FMBF) is developed as a noise reduction method. It is based on an IR imaging mechanism to detect the noisy pixels and on a modified median-based filter to remove the noisy pixels in IR images. FMBF has the advantage of a low computation load. In addition, FMBF can retain reasonably good edges and texture information when the size of the filter window increases. The most important advantage is that the peak signal-to-noise ratio (PSNR) caused by FMBF is higher than the PSNR caused by the median filter. A hybrid cumulative histogram equalization (HCHE) is proposed for adaptive contrast enhancement. HCHE can automatically generate a hybrid cumulative histogram (HCH) based on two different pieces of information about the image histogram. HCHE can improve the enhancement effect on hot objects rather than background. The experimental results are addressed and demonstrate that the proposed approach is feasible for use as an effective and adaptive process for enhancing the quality of IR vein-pattern images.

The comparison of radiation dose between C-arm flat-detector CT (DynaCT) and multi-slice CT (MSCT): A phantom study

PurposeTo assess and compare the radiation dose of DynaCT in phantoms to that of MSCT. Methods and materialsA male Anderson Radiation Therapy 200 phantom with embedded thermoluminescence dosimeters (TLDs) was scanned with DynaCT (Axiom Artis dTA, Siemens Healthcare, Forchheim, Germany) and MSCTs (MSCT1: Lightspeed VCT, GE, Milwaukee, USA; MSCT2: Sensation Cardiac 64, Siemens Healthcare, Erlangen, Germany), respectively. For DynaCT, radiation exposure data with automatic exposure control were obtained from scanning of head, chest and abdomen. For MSCT, scanning protocols for head, chest, and abdomen were examined under conventional exposure conditions. Effective doses were calculated from LiF–TLD measurements according to ICRP103. A Catphan phantom (Phantom Lab, USA) was used to evaluate the spatial resolution and low contrast detectability of DynaCT and MSCT. ResultsThe effective doses for DynaCT from head, chest and abdomen scanning were 1.18, 7.32, 7.48mSv (20s scan) and 0.85, 6.01, 7.04mSv (8s scan); the effective doses for MSCT from head, chest and abdomen scanning were 3.33, 7.62 and 8.42mSv (MSCT1), and 1.89, 7.52, 8.23mSv (MSCT2). Significant difference between the organ doses from DynaCT and from MSCT (p<.05) was shown. The spatial resolution of 12lp/cm was achieved and it was able to recognize a 3mm low contrast object at 0.5% contrast level in DynaCT, which was on the same level as in the MSCT images. ConclusionThe phantom study indicates that DynaCT applies significantly less dose to patient and achieves similar spatial resolution and low contrast detectability to standard diagnostic MSCT.

The Accuracy of Drivers' Judgments of the Effects of Headlight Glare on the Visibility of a Non-retroreflective Object

Headlights must balance the conflicting goals of maximizing visibility for the driver and minimizing glare to others. Yet consumer complaints about headlights tend to emphasize only glare. This project extended earlier research by asking 17 young drivers to judge the distance at which they would be able to recognize the orientation of a non-retroreflective stimulus adjacent to the headlights of a stationary opposing vehicle. The beam setting (low vs. high) of both vehicles was manipulated. Participants correctly estimated that the stimulus would be less visible when their own vehicle used low beams and when the opposing vehicle used high beams. Contrary to earlier studies that used high contrast (retroreflective) stimuli, drivers' estimates of conspicuity were, on average, not significantly different from the actual visibility of the stimulus. These results suggest that drivers' understanding of headlight glare is more accurate for low contrast stimuli than for high contrast (retroreflective) stimuli and that drivers may overestimate the disabling effects of opposing headlights for higher contrast objects more than for lower contrast objects. Language: en

Sparsity and low‐contrast object detectability

The application of sparsity-driven reconstruction methods to MRI to date has largely focused on situations where high-contrast features (e.g., gadolinium-enhanced vessels) are of primary interest. In clinical practice, however, low contrast features such as subtle lesions are often of equal or greater interest. Using an American College of Radiology MR quality assurance phantom and test, we describe a novel framework for systematically and automatically evaluating the low-contrast object detectability performance of different undersampled image reconstruction methods. This platform is used to evaluate three such methods, two based on classic Tikhonov regularization and one sparsity-driven method based on ℓ(1) -norm minimization (which is commonly used in compressive sensing, also known as compressed sensing, applications), across a wide range of sampling rates and parameterizations. Both the automated evaluation system and a manual evaluation of anatomical images with numerically-generated low contrast inserts demonstrate that sparse reconstructions exhibit superior low-contrast object detectability performance compared to both Tikhonov-regularized reconstructions. The implications of this result, and potential applications of both the described low-contrast object detectability platform and generalizations of it are then discussed.

Effects of the penalty on the penalized weighted least-squares image reconstruction for low-dose CBCT

Statistical iterative reconstruction (SIR) algorithms have shown potential to substantially improve low-dose cone-beam CT (CBCT) image quality. The penalty term plays an important role in determining the performance of SIR algorithms. In this work, we quantitatively evaluate the impact of the penalties on the performance of a statistics-based penalized weighted least-squares (PWLS) iterative reconstruction algorithm for improving the image quality of low-dose CBCT. Three different edge-preserving penalty terms, exponential form anisotropic quadratic (AQ) penalty (PWLS-Exp), inverse square form AQ penalty (PWLS-InverseSqr) and total variation penalty (PWLS-TV), were compared against the conventional isotropic quadratic form penalty (PWLS-Iso) using both computer simulation and experimental studies. Noise in low-dose CBCT can be substantially suppressed by the PWLS reconstruction algorithm and edges are well preserved by both AQ- and TV-based penalty terms. The noise-resolution tradeoff measurement shows that the PWLS-Exp exhibits the best spatial resolution of all the three anisotropic penalty terms at matched noise level for reconstructing high-contrast objects. For the reconstruction of low-contrast objects, the TV-based penalty outperforms the AQ-based one with better resolution preservation at matched noise levels. Different penalty terms may be used for better edge preservation at different targeted contrast levels.

Automated assessment of low contrast sensitivity for CT systems using a model observer

Low contrast sensitivity of CT scanners is regularly assessed by subjective scoring of low contrast detectability within phantom CT images. Since in these phantoms low contrast objects are arranged in known fixed patterns, subjective rating of low contrast visibility might be biased. The purpose of this study was to develop and validate a software for automated objective low contrast detectability based on a model observer. Images of the low contrast module of the Catphan 600 phantom were used for the evaluation of the software. This module contains two subregions: the supraslice region with three groups of low contrast objects (each consisting of nine circular objects with diameter 2-15 mm and contrast 0.3, 0.5, and 1.0%, respectively) and the subslice region with three groups of four circular objects each (diameter 3-9 mm; contrast 1.0%). The software method offered automated determination of low contrast detectability using a NPWE (nonprewhitening matched filter with an eye filter) model observer for the supraslice region. The model observer correlated templates of the low contrast objects with the acquired images of the Catphan phantom and a discrimination index d' was calculated. This index was transformed into a proportion correct (PC) value. In the two-alternative forced choice (2-AFC) experiments used in this study, a PC ≥ 75% was proposed as a threshold to decide whether objects were visible. As a proof of concept, influence of kVp (between 80 and 135 kV), mAs (25-200 mAs range) and reconstruction filter (four filters, two soft and two sharp) on low contrast detectability was investigated. To validate the outcome of the software in a qualitative way, a human observer study was performed. The expected influence of kV, mAs and reconstruction filter on image quality are consistent with the results of the proposed automated model. Higher values for d' (or PC) are found with increasing mAs or kV values and for the soft reconstruction filters. For the highest contrast group (1%), PC values were fairly above 75% for all object diameters >2 mm, for all conditions. For the 0.5% contrast group, the same behavior was observed for object diameters >3 mm for all conditions. For the 0.3% contrast group, PC values were higher than 75% for object diameters >6 mm except for the series acquired at the lowest dose (25 mAs), which gave lower PC values. In the human observer study similar trends were found. We have developed an automated method to objectively investigate image quality using the NPWE model in combination with images of the Catphan phantom low contrast module. As a first step, low contrast detectability as a function of both acquisition and reconstruction parameter settings was successfully investigated with the software. In future work, this method could play a role in image reconstruction algorithms evaluation, dose reduction strategies or novel CT technologies, and other model observers may be implemented as well.

A segmentation method to obtain a complete geometry model of the hearing organ

We present a method for obtaining a complete geometry model of the fluid chambers of cochlea ( scalae) from tomography images. An accurate segmentation of cochlea is problematic due to the low contrast of the inner membranes of scalae. Our method of 3D segmentation is based on dynamic resampling of an original image stack to achieve a perpendicular cross-section of the scalae on all sections. Subsequently, perpendicular cross-section is being segmented using 2D active contours. The center of mass of the contour is extracted and used to predict further course of scalae centerline by Kalman filter. Cross-section contours are subsequently assembled to the total geometry model. This method has been applied to CT images, but we expect that it could be used for segmentation of strongly curved low-contrast tubular objects recorded with other tomography techniques.