Accuracy and reproducibility of fast fractional flow reserve computation from invasive coronary angiography.

Fractional flow reserve (FFR) guided percutaneous coronary intervention (PCI) is associated with favourable outcome compared with revascularization based on angiographic stenosis severity alone. The feasibility of the new image-based quantitative flow ratio (QFR) assessed from 3D quantitative coronary angiography (QCA) and thrombolysis in myocardial infarction (TIMI) frame count using three different flow models has been reported recently. The aim of the current study was to assess the accuracy, and in particular, the reproducibility of these three QFR techniques when compared with invasive FFR. QFR was derived (1) from adenosine induced hyperaemic coronary angiography images (adenosine-flow QFR [aQFR]), (2) from non-hyperemic images (contrast-flow QFR [cQFR]) and (3) using a fixed empiric hyperaemic flow [fixed-flow QFR (fQFR)]. The three QFR values were calculated in 17 patients who prospectively underwent invasive FFR measurement in 20 vessels. Two independent observers performed the QFR analyses. Mean difference, standard deviation and 95% limits of agreement (LOA) between invasive FFR and aQFR, cQFR and fQFR for observer 1 were: 0.01 ± 0.04 (95% LOA: -0.07; 0.10), 0.01 ± 0.05 (95% LOA: -0.08; 0.10), 0.01 ± 0.04 (95% LOA: -0.06; 0.08) and for observer 2: 0.00 ± 0.03 (95% LOA: -0.06; 0.07), -0.01 ± 0.03 (95% LOA: -0.07; 0.05), 0.00 ± 0.03 (95% LOA: -0.06; 0.05). Values between the 2 observers were (to assess reproducibility) for aQFR: 0.01 ± 0.04 (95% LOA: -0.07; 0.09), for cQFR: 0.02 ± 0.04 (95% LOA: -0.06; 0.09) and for fQFR: 0.01 ± 0.05 (95% LOA: -0.07; 0.10). In a small number of patients we showed good accuracy of three QFR techniques (aQFR, cQFR and fQFR) to predict invasive FFR. Furthermore, good inter-observer agreement of the QFR values was observed between two independent observers.

Towards local progression estimation of pulmonary emphysema using CT.

PURPOSE: Whole lung densitometry on chest CT images is an accepted method for measuring tissue destruction in patients with pulmonary emphysema in clinical trials. Progression measurement is required for evaluation of change in health condition and the effect of drug treatment. Information about the location of emphysema progression within the lung may be important for the correct interpretation of drug efficacy, or for determining a treatment plan. The purpose of this study is therefore to develop and validate methods that enable the local measurement of lung density changes, which requires proper modeling of the effect of respiration on density. METHODS: Four methods, all based on registration of baseline and follow-up chest CT scans, are compared. The first naïve method subtracts registered images. The second employs the so-called dry sponge model, where volume correction is performed using the determinant of the Jacobian of the transformation. The third and the fourth introduce a novel adaptation of the dry sponge model that circumvents its constant-mass assumption, which is shown to be invalid. The latter two methods require a third CT scan at a different inspiration level to estimate the patient-specific density-volume slope, where one method employs a global and the other a local slope. The methods were validated on CT scans of a phantom mimicking the lung, where mass and volume could be controlled. In addition, validation was performed on data of 21 patients with pulmonary emphysema. RESULTS: The image registration method was optimized leaving a registration error below half the slice increment (median 1.0 mm). The phantom study showed that the locally adapted slope model most accurately measured local progression. The systematic error in estimating progression, as measured on the phantom data, was below 2 gr/l for a 70 ml (6%) volume difference, and 5 gr/l for a 210 ml (19%) difference, if volume correction was applied. On the patient data an underlying linearity assumption relating lung volume change with density change was shown to hold (fitR(2) = 0.94), and globalized versions of the local models are consistent with global results (R(2) of 0.865 and 0.882 for the two adapted slope models, respectively). CONCLUSIONS: In conclusion, image matching and subsequent analysis of differences according to the proposed lung models (i) has good local registration accuracy on patient data, (ii) effectively eliminates a dependency on inspiration level at acquisition time, (iii) accurately predicts progression in phantom data, and (iv) is reasonably consistent with global results in patient data. It is therefore a potential future tool for assessing local emphysema progression in drug evaluation trials and in clinical practice.

Automated registration of multispectral MR vessel wall images of the carotid artery.

PURPOSE: Atherosclerosis is the primary cause of heart disease and stroke. The detailed assessment of atherosclerosis of the carotid artery requires high resolution imaging of the vessel wall using multiple MR sequences with different contrast weightings. These images allow manual or automated classification of plaque components inside the vessel wall. Automated classification requires all sequences to be in alignment, which is hampered by patient motion. In clinical practice, correction of this motion is performed manually. Previous studies applied automated image registration to correct for motion using only nondeformable transformation models and did not perform a detailed quantitative validation. The purpose of this study is to develop an automated accurate 3D registration method, and to extensively validate this method on a large set of patient data. In addition, the authors quantified patient motion during scanning to investigate the need for correction. METHODS: MR imaging studies (1.5T, dedicated carotid surface coil, Philips) from 55 TIA∕stroke patients with ipsilateral <70% carotid artery stenosis were randomly selected from a larger cohort. Five MR pulse sequences were acquired around the carotid bifurcation, each containing nine transverse slices: T1-weighted turbo field echo, time of flight, T2-weighted turbo spin-echo, and pre- and postcontrast T1-weighted turbo spin-echo images (T1W TSE). The images were manually segmented by delineating the lumen contour in each vessel wall sequence and were manually aligned by applying throughplane and inplane translations to the images. To find the optimal automatic image registration method, different masks, choice of the fixed image, different types of the mutual information image similarity metric, and transformation models including 3D deformable transformation models, were evaluated. Evaluation of the automatic registration results was performed by comparing the lumen segmentations of the fixed image and moving image after registration. RESULTS: The average required manual translation per image slice was 1.33 mm. Translations were larger as the patient was longer inside the scanner. Manual alignment took 187.5 s per patient resulting in a mean surface distance of 0.271 ± 0.127 mm. After minimal user interaction to generate the mask in the fixed image, the remaining sequences are automatically registered with a computation time of 52.0 s per patient. The optimal registration strategy used a circular mask with a diameter of 10 mm, a 3D B-spline transformation model with a control point spacing of 15 mm, mutual information as image similarity metric, and the precontrast T1W TSE as fixed image. A mean surface distance of 0.288 ± 0.128 mm was obtained with these settings, which is very close to the accuracy of the manual alignment procedure. The exact registration parameters and software were made publicly available. CONCLUSIONS: An automated registration method was developed and optimized, only needing two mouse clicks to mark the start and end point of the artery. Validation on a large group of patients showed that automated image registration has similar accuracy as the manual alignment procedure, substantially reducing the amount of user interactions needed, and is multiple times faster. In conclusion, the authors believe that the proposed automated method can replace the current manual procedure, thereby reducing the time to analyze the images.

Mixed integer evolution strategies for parameter optimization.

Evolution strategies (ESs) are powerful probabilistic search and optimization algorithms gleaned from biological evolution theory. They have been successfully applied to a wide range of real world applications. The modern ESs are mainly designed for solving continuous parameter optimization problems. Their ability to adapt the parameters of the multivariate normal distribution used for mutation during the optimization run makes them well suited for this domain. In this article we describe and study mixed integer evolution strategies (MIES), which are natural extensions of ES for mixed integer optimization problems. MIES can deal with parameter vectors consisting not only of continuous variables but also with nominal discrete and integer variables. Following the design principles of the canonical evolution strategies, they use specialized mutation operators tailored for the aforementioned mixed parameter classes. For each type of variable, the choice of mutation operators is governed by a natural metric for this variable type, maximal entropy, and symmetry considerations. All distributions used for mutation can be controlled in their shape by means of scaling parameters, allowing self-adaptation to be implemented. After introducing and motivating the conceptual design of the MIES, we study the optimality of the self-adaptation of step sizes and mutation rates on a generalized (weighted) sphere model. Moreover, we prove global convergence of the MIES on a very general class of problems. The remainder of the article is devoted to performance studies on artificial landscapes (barrier functions and mixed integer NK landscapes), and a case study in the optimization of medical image analysis systems. In addition, we show that with proper constraint handling techniques, MIES can also be applied to classical mixed integer nonlinear programming problems.

Automated versus manual in vivo segmentation of carotid plaque MRI.

BACKGROUND AND PURPOSE: Automatically identifying carotid plaque composition using MR imaging remains a challenging task in vivo. The purpose of our study was to compare the detection and quantification of carotid artery atherosclerotic plaque components based on in vivo MR imaging data using manual and automated segmentation. MATERIALS AND METHODS: Sixty patients from a multicenter study were split into a training group (20 patients) and a study group (40 patients). Each MR imaging study consisted of 4 high-resolution carotid wall sequences (T1, T2, PDw, TOF). Manual segmentation was performed by delineation of the vessel wall and different plaque components. Automated segmentation was performed in the study group by a supervised classifier trained on images from the training group of patients. RESULTS: For the detection of plaque components, the agreement between the visual and automated analysis was moderate for calcifications (κ = 0.59, CI 95% [0.36-0.82]) and good for hemorrhage (0.65 [0.42-0.88]) and lipids (0.65 [0.03-1.27]). For quantification of plaque volumes, the intraclass correlation was high for hemorrhage (0.80 [0.54-0.92]) and fibrous tissue (0.80 [0.65-0.89]), good for lipids (0.65 [0.43-0.80]), and poor for calcifications. CONCLUSIONS: In 40 patients with carotid stenosis, our results indicated that it was possible to automatically detect carotid plaque components with substantial or good agreement with visual identification, and that the volumes obtained manually and automatically were reasonably consistent for hemorrhage and lipids but not for calcium.

Non-invasive assessment of atherosclerotic coronary lesion length using multidetector computed tomography angiography: comparison to quantitative coronary angiography.

Multidetector computed tomography angiography (CTA) provides information on plaque extent and stenosis in the coronary wall. More accurate lesion assessment may be feasible with CTA as compared to invasive coronary angiography (ICA). Accordingly, lesion length assessment was compared between ICA and CTA in patients referred for CTA who underwent subsequent percutaneous coronary intervention (PCI). 89 patients clinically referred for CTA were subsequently referred for ICA and PCI. On CTA, lesion length was measured from the proximal to the distal shoulder of the plaque. Quantitative coronary angiography (QCA) was performed to analyze lesion length. Stent length was recorded for each lesion. In total, 119 lesions were retrospectively identified. Mean lesion length on CTA was 21.4 ± 8.4 mm and on QCA 12.6 ± 6.1 mm. Mean stent length deployed was 17.4 ± 5.3 mm. Lesion length on CTA was significantly longer than on QCA (difference 8.8 ± 6.7 mm, P < 0.001). Moreover, lesion length visualized on CTA was also significantly longer than mean stent length (CTA lesion length-stent length was 4.2 ± 8.7 mm, P < 0.001). Lesion length assessed by CTA is longer than that assessed by ICA. Possibly, CTA provides more accurate lesion length assessment than ICA and may facilitate improved guidance of percutaneous treatment of coronary lesions.

Performance and efficacy of 320-row computed tomography coronary angiography in patients presenting with acute chest pain: results from a clinical registry.

The purpose of this study was to evaluate the performance of 320-row computed tomography angiography (CTA) in the identification of significant coronary artery disease (CAD) in patients presenting with acute chest pain and to examine the relation to outcome during follow-up. A total of 106 patients with acute chest pain underwent CTA to evaluate presence of CAD. Each CTA was classified as: normal, non-significant CAD (<50% luminal narrowing) and significant CAD (≥50% luminal narrowing). CTA results were compared with quantitative coronary angiography. After discharge, the following cardiovascular events were recorded: cardiac death, non-fatal infarction, and unstable angina requiring revascularization. Among the 106 patients, 23 patients (22%) had a normal CTA, 19 patients (18%) had non-significant CAD on CTA, 59 patients (55%) had significant CAD on CTA, and 5 patients (5%) had non-diagnostic image quality. In total, 16 patients (15%) were immediately discharged after normal CTA and 90 patients (85%) underwent invasive coronary angiography. Sensitivity, specificity, and positive and negative predictive values to detect significant CAD on CTA were 100, 87, 93, and 100%, respectively. During mean follow-up of 13.7 months, no cardiovascular events occurred in patients with a normal CTA examination. In patients with non-significant CAD on CTA, no cardiac death or myocardial infarctions occurred and only 1 patient underwent revascularization due to unstable angina. In patients presenting with acute chest pain, an excellent clinical performance for the non-invasive assessment of significant CAD was demonstrated using CTA. Importantly, normal or non-significant CAD on CTA predicted a low rate of adverse cardiovascular events and favorable outcome during follow-up.

Comprehensive visualization of multimodal cardiac imaging data for assessment of coronary artery disease: first clinical results of the SMARTVis tool.

PURPOSE: In clinical practice, both coronary anatomy and myocardial perfusion information are needed to assess coronary artery disease (CAD). The extent and severity of coronary stenoses can be determined using computed tomography coronary angiography (CTCA); the presence and amount of ischemia can be identified using myocardial perfusion imaging, such as perfusion magnetic resonance imaging (PMR). To determine which specific stenosis is associated with which ischemic region, experts use assumptions on coronary perfusion territories. Due to the high variability between patient's coronary artery anatomies, as well as the uncertain relation between perfusion territories and supplying coronary arteries, patient-specific systems are needed. MATERIAL AND METHODS: We present a patient-specific visualization system, called Synchronized Multimodal heART Visualization (SMARTVis), for relating coronary stenoses and perfusion deficits derived from CTCA and PMR, respectively. The system consists of the following comprehensive components: (1) two or three-dimensional fusion of anatomical and functional information, (2) automatic detection and ranking of coronary stenoses, (3) estimation of patient-specific coronary perfusion territories. RESULTS: The potential benefits of the SMARTVis tool in assessing CAD were investigated through a case-study evaluation (conventional vs. SMARTVis tool): two experts analyzed four cases of patients with suspected multivessel coronary artery disease. When using the SMARTVis tool, a more reliable estimation of the relation between perfusion deficits and stenoses led to a more accurate diagnosis, as well as a better interobserver diagnosis agreement. CONCLUSION: The SMARTVis comprehensive visualization system can be effectively used to assess disease status in multivessel CAD patients, offering valuable new options for the diagnosis and management of these patients.

Integrated contour detection and pose estimation for fluoroscopic analysis of knee implants.

With fluoroscopic analysis of knee implant kinematics the implant contour must be detected in each image frame, followed by estimation of the implant pose. With a large number of possibly low-quality images, the contour detection is a time-consuming bottleneck. The present paper proposes an automated contour detection method, which is integrated in the pose estimation. In a phantom experiment the automated method was compared with a standard method, which uses manual selection of correct contour parts. Both methods demonstrated comparable precision, with a minor difference in the Y-position (0.08 mm versus 0.06 mm). The precision of each method was so small (below 0.2 mm and 0.3 degrees) that both are sufficiently accurate for clinical research purposes. The efficiency of both methods was assessed on six clinical datasets. With the automated method the observer spent 1.5 min per image, significantly less than 3.9 min with the standard method. A Bland-Altman analysis between the methods demonstrated no discernible trends in the relative femoral poses. The threefold increase in efficiency demonstrates that a pose estimation approach with integrated contour detection is more intuitive than a standard method. It eliminates most of the manual work in fluoroscopic analysis, with sufficient precision for clinical research purposes.

The robustness and accuracy of in vivo linear wear measurements for knee prostheses based on model-based RSA.

Accurate in vivo measurements methods of wear in total knee arthroplasty are required for a timely detection of excessive wear and to assess new implant designs. Component separation measurements based on model-based Roentgen stereophotogrammetric analysis (RSA), in which 3-dimensional reconstruction methods are used, have shown promising results, yet the robustness of these measurements is unknown. In this study, the accuracy and robustness of this measurement for clinical usage was assessed. The validation experiments were conducted in an RSA setup with a phantom setup of a knee in a vertical orientation. 72 RSA images were created using different variables for knee orientations, two prosthesis types (fixed-bearing Duracon knee and fixed-bearing Triathlon knee) and accuracies of the reconstruction models. The measurement error was determined for absolute and relative measurements and the effect of knee positioning and true seperation distance was determined. The measurement method overestimated the separation distance with 0.1mm on average. The precision of the method was 0.10mm (2*SD) for the Duracon prosthesis and 0.20mm for the Triathlon prosthesis. A slight difference in error was found between the measurements with 0° and 10° anterior tilt. (difference=0.08mm, p=0.04). The accuracy of 0.1mm and precision of 0.2mm can be achieved for linear wear measurements based on model-based RSA, which is more than adequate for clinical applications. The measurement is robust in clinical settings. Although anterior tilt seems to influence the measurement, the size of this influence is low and clinically irrelevant.

Automated analysis of three-dimensional stress echocardiography.

Real-time three-dimensional (3D) ultrasound imaging has been proposed as an alternative for two-dimensional stress echocardiography for assessing myocardial dysfunction and underlying coronary artery disease. Analysis of 3D stress echocardiography is no simple task and requires considerable expertise. In this paper, we propose methods for automated analysis, which may provide a more objective and accurate diagnosis. Expert knowledge is incorporated via statistical modelling of patient data. Methods for identifying anatomical views, detecting endocardial borders, and classification of wall motion are described and shown to provide favourable results. We also present software developed especially for analysis of 3D stress echocardiography in clinical practice. Interobserver agreement in wall motion scoring is better using the dedicated software (96%) than commercially available software not dedicated for this purpose (79%). The developed tools may provide useful quantitative and objective parameters to assist the clinical expert in the diagnosis of left ventricular function.

Interactive navigation of segmented MR angiograms using simultaneous curved planar and volume visualizations.

PURPOSE: Interactive visualization is required to inspect and monitor the automatic segmentation of vessels derived from contrast-enhanced magnetic resonance angiography (CE-MRA). A dual-view visualization scheme consisting of curved planar reformation (CPR) and direct volume rendering (DVR) was developed for this purpose and tested. METHODS: A dual view visualization scheme was developed using the vessel pathline for both camera position and rotation in 3D, greatly reducing the degrees of freedom (DOF) required for navigation. Pathline-based navigation facilitates coupling of the CPR and DVR views, as local position and orientation can be matched precisely. The new technique was compared to traditional techniques in a user study. Layperson users were required to perform a visual search task that involves checking for (minor) errors in segmentations of MRA data from a software phantom. The task requires the user to examine both views. RESULTS: Pathline-based navigation and coupling of CPR and DVR provide user speed performance improvements in a vessel inspection task. Interactive MRA visualization with this method, where rotational degrees of freedom were reduced, had no negative effect. CONCLUSIONS: The DOF reduction achieved by the new navigation technique is beneficial to user performance. The technique is promising and merits comprehensive evaluation in a realistic clinical setting.

Neurological signs in relation to white matter hyperintensity volumes in memory clinic patients.

PURPOSE: To determine the frequency of neurological signs in a memory clinic population and to explore their associations with white matter hyperintensity (WMH). METHODS: We included patients with Alzheimer disease (AD; n = 210), vascular dementia (VaD; n = 34), mild cognitive impairment (MCI; n = 86) and subjective complaints (n = 153). The presence of extrapyramidal and unilateral signs was assessed from medical charts. On MRI, WMH volumes were extracted automatically. RESULTS: Extrapyramidal signs were found in 10% and unilateral signs in 12% of the patients. Age- and sex-adjusted extrapyramidal signs occurred more often in VaD compared to patients with subjective complaints. Unilateral signs were more prevalent in all groups compared to patients with subjective complaints. Two-way analysis of variance (ANOVA) with WMH as the dependent variable showed a main effect of diagnosis (p < 0.001), but not of extrapyramidal signs (p = 0.62). In contrast, 2-way ANOVA showed main effects of diagnosis (p < 0.001) and unilateral signs (p = 0.001). Furthermore, there was an interaction between these factors (p = 0.04); if unilateral signs were present, patients with subjective complaints and VaD showed more WMH, whereas there was no relation in AD and MCI. CONCLUSION: Extrapyramidal and unilateral signs are common in memory clinic patients, but are only modestly related to WMH.

Detecting femur-insert collisions to improve precision of fluoroscopic knee arthroplasty analysis.

Fluoroscopic analysis is an important tool for assessing in vivo kinematics of knee prostheses. Most commonly, a single-plane fluoroscopic setup is used to capture the motion of prostheses during a particular task. Unfortunately, single-plane fluoroscopic analysis is imprecise in the out-of-plane direction. This can result in reconstructing physically impossible poses, in which--for example--the femoral component intersects with the insert, as the normal pose estimation process does not take into account the relation between the components. In the proposed method, the poses of both components are estimated simultaneously, while preventing femur-insert collisions. In a phantom study, the accuracy and precision of the new method in estimating the relative pose of the femoral component were compared to those of the original method. With reverse engineered models, the errors in estimating the out-of-plane position decreased from 2.0+/-0.7 to 0.1+/-0.1 mm, without effects on the errors in rotations and the in-plane positions. With CAD models, the errors in estimating the out-of-plane position decreased from 5.3+/-0.7 mm (mean+/-SD) to 0.0+/-0.4 mm, at the expense of a decreased precision for the other position or orientation parameters. In conclusion, collision detection can prevent reconstructing impossible poses and it improves the position and motion estimation in the out-of-plane direction.

Plaque type and composition as evaluated non-invasively by MSCT angiography and invasively by VH IVUS in relation to the degree of stenosis.

BACKGROUND: Imaging of coronary plaques has traditionally focused on evaluating degree of stenosis, as the risk for adverse cardiac events increases with stenosis severity. However, the relation between plaque composition and severity of stenosis remains largely unknown. OBJECTIVE: To assess plaque composition (non-invasively by multislice computed tomography (MSCT) angiography and invasively by virtual histology intravascular ultrasound (VH IVUS)) in relation to degree of stenosis. METHODS: 78 patients underwent MSCT (identifying three plaque types; non-calcified, calcified, mixed) followed by invasive coronary angiography and VH IVUS. VH IVUS evaluated plaque burden, minimal lumen area and plaque composition (fibrotic, fibro-fatty, necrotic core, dense calcium) and plaques were classified as fibrocalcific, fibroatheroma, thin-capped fibroatheroma (TCFA), pathological intimal thickening. For each plaque, percentage stenosis was evaluated by quantitative coronary angiography. Significant stenosis was defined >50% stenosis. RESULTS: Overall, 43 plaques (19%) corresponded to significant stenosis. Of the 227 plaques analysed, 70 were non-calcified plaques (31%), 96 mixed (42%) and 61 calcified (27%) on MSCT. Plaque types on MSCT were equally distributed among significant and non-significant stenoses. VH IVUS identified that plaques with significant stenosis had higher plaque burden (67% (11%) vs 53% (12%), p<0.05) and smaller minimal lumen area (4.6 (3.8-6.8) mm(2) vs 7.3 (5.4-10.5) mm(2), p<0.05). Interestingly, no differences were observed in percentage fibrotic, fibro-fatty, necrotic core and dense calcium. Non-significant stenoses were more frequently classified as pathological intimal thickening (46 (25%) vs 3 (7%), p<0.05), although TCFA (more vulnerable plaque) was distributed equally (p = 0.18). CONCLUSION: No evident association exists between the degree of stenosis and plaque composition or vulnerability, as evaluated non-invasively by MSCT and invasively by VH IVUS.

Prospects for early detection of Alzheimer's disease from serial MR images in transgenic mice.

The existing literature on the magnetic resonance imaging of murine models of Alzheimer's disease is reviewed. Particular attention is paid to the possibilities for the early detection of the disease. To this effect, not only are relaxometric and volumetric approaches discussed, but also mathematical models for plaque distribution and aggregation. Image analysis plays a prominent role in this line of research, as stochastic image models and texture analysis have shown some success in the classification of subjects affected by Alzheimer's disease. It is concluded that relaxometric approaches seem to be a promising candidate for the task at hand, especially when combined with sophisticated image analysis, and when data from more than one time-point is available. There have been few longitudinal studies of mice models so far, so this direction of research warrants future efforts.

User-agent cooperation in multiagent IVUS image segmentation.

Automated interpretation of complex images requires elaborate knowledge and model-based image analysis, but often needs interaction with an expert as well. This research describes expert interaction with a multiagent image interpretation system using only a restricted vocabulary of high-level user interactions. The aim is to minimize inter- and intra-observer variability by keeping the total number of interactions as low and simple as possible. The multiagent image interpretation system has elaborate high-level knowledge-based control over low-level image segmentation algorithms. Agents use contextual knowledge to keep the number of interactions low but, when in doubt, present the user with the most likely interpretation of the situation. The user, in turn, can correct, supplement, and/or confirm the results of image-processing agents. This is done at a very high level of abstraction such that no knowledge of the underlying segmentation methods, parameters or agent functioning is needed. High-level interaction thereby replaces more traditional contour correction methods like inserting points and/or (re)drawing contours. This makes it easier for the user to obtain good results, while inter- and intra-observer variability are kept minimal, since the image segmentation itself remains under control of image-processing agents. The system has been applied to intravascular ultrasound (IVUS) images. Experiments show that with an average of 2-3 high-level user interactions per correction, segmentation results substantially improve while the variation is greatly reduced. The achieved level of accuracy and repeatability is equivalent to that of manual drawing by an expert.