Rolf A. Heckemann

Automatic anatomical brain MRI segmentation combining label propagation and decision fusion.

Regions in three-dimensional magnetic resonance (MR) brain images can be classified using protocols for manually segmenting and labeling structures. For large cohorts, time and expertise requirements make this approach impractical. To achieve automation, an individual segmentation can be propagated to another individual using an anatomical correspondence estimate relating the atlas image to the target image. The accuracy of the resulting target labeling has been limited but can potentially be improved by combining multiple segmentations using decision fusion. We studied segmentation propagation and decision fusion on 30 normal brain MR images, which had been manually segmented into 67 structures. Correspondence estimates were established by nonrigid registration using free-form deformations. Both direct label propagation and an indirect approach were tested. Individual propagations showed an average similarity index (SI) of 0.754+/-0.016 against manual segmentations. Decision fusion using 29 input segmentations increased SI to 0.836+/-0.009. For indirect propagation of a single source via 27 intermediate images, SI was 0.779+/-0.013. We also studied the effect of the decision fusion procedure using a numerical simulation with synthetic input data. The results helped to formulate a model that predicts the quality improvement of fused brain segmentations based on the number of individual propagated segmentations combined. We demonstrate a practicable procedure that exceeds the accuracy of previous automatic methods and can compete with manual delineations.

Pulse-inversion mode imaging of liver specific microbubbles : improved detection of subcentimetre metastases

Pulse-inversion mode (a new ultrasound mode) can be used to image the late liver-specific parenchymal phase of the microbubble contrast-agent Levovist. Scanning in pulse-inversion mode after Levovist improves the detection of liver metastases and reveals more lesions of smaller size than conventional ultrasonography and computed tomography.

Liver microbubble transit time compared with histology and Child-Pugh score in diffuse liver disease: a cross sectional study.

Background: A previous pilot study showed that early arrival time of a microbubble in a hepatic vein is a sensitive indicator of cirrhosis. Aim: To see if this index can also grade diffuse liver disease. Patients: Thirty nine fasted patients with histologically characterised disease were studied prospectively. Nine patients had no evidence of liver fibrosis, 10 had fibrosis without cirrhosis, and 20 had cirrhosis (five Child’s A, seven Child’s B, and eight Child’s C). Methods: Bolus injections of a microbubble (Levovist; Schering, Berlin) were given intravenously, followed by a saline flush. Time intensity curves of hepatic vein and carotid artery spectral Doppler signals were analysed. Hepatic vein transit time (HVTT) was calculated as the time after injection at which a sustained signal increase >10% of baseline was seen. Carotid delay time (CDT) was calculated as the difference between carotid and hepatic vein enhancement. Results: Diagnostic studies were achieved in 38/39 subjects. Both HVTT and CDT became consistently shorter with worsening disease, as follows (means (SD)): HVTT: no fibrosis 44 (25) s, fibrosis 26 (8) s, Child’s A 21 (1) s, Child’s B 16 (3) s, and Child’s C 16 (2) s; CDT: no fibrosis 31 (29) s, fibrosis 14 (6) s, Child’s A 8 (1) s, Child’s B 4 (4) s, and Child’s C 3 (3) s. These differences were highly significant (p<0.001, ANOVA comparison). A HVTT <24 s and a CDT <10 s were 100% sensitive for cirrhosis (20/20 and 18/18, respectively) but not completely specific: 2/8 subjects with fibrosis had CDT values <10 s and 3/9 had HVTT <24 s. Conclusion: This minimally invasive test shows promise not only in diagnosing cirrhosis but also in assessing disease severity.

Stimulated acoustic emission to image a late liver and spleen-specific phase of Levovist® in normal volunteers and patients with and without liver disease

Quantitative studies were performed to investigate liver- specific uptake of the microbubble Levovist, using stimulated acoustic emission (SAE), which can detect microbubbles even when stationary or slow-moving. These comprised studies of biodistribution comparing the liver and kidney in five normal volunteers, reproducibility in 34 patients, comparison between cirrhotics and controls (n = 9 each) and maximal depth of effect at different frequencies (180 measurements in 31 patients). Stimulated acoustic emission lasted beyond 30 min, with strongly liver-specific properties in each volunteer and was highly reproducible. No difference in the amount of SAE in the superficial liver was seen between cirrhotic and normal livers, but attenuation was higher in cirrhotics. This demonstrates a frequency-dependent effect on liver SAE penetration. We conclude that the liver uptake of Levovist lasts over 30 min, is reproducible, occurs even where diffuse liver disease is present and can be used to assess tissue attenuation in a novel fashion.

COMPARE radiology: Creating an interactive web-based training program for radiology with multimedia authoring software

RATIONALE AND OBJECTIVESnComputer-based training has two primary benefits: Content can be presented interactively, and students can choose the time, place, and pace of learning. As a subject of medical education, radiology lends itself particularly well to computer-based training because of its highly visual content. To improve the efficiency of radiology training at their institution, the authors decided to create an interactive Web-based training site.nnnMATERIALS AND METHODSnWorking with a group of medical students knowledgeable in multimedia authoring, the authors used authoring software to create COMPARE Radiology, an interactive training program that follows the modality-based structure of the undergraduate curriculum for radiology at the University of Erlangen-Nuremberg, Erlangen, Germany, and at medical schools worldwide.nnnRESULTSnThe Web-based program offers cases and exercises in radiographic anatomy at different selectable levels of difficulty, allowing users to test and build their knowledge of radiology. Pathologic images are initially presented without any further information. Additional information (patient history, laboratory results, reports from other imaging studies, and normal images for comparison) can be retrieved selectively and successively. Further information regarding the diagnosis and pathologic findings can be found by following links to external Web sites. The COMPARE Radiology program content is extended and updated regularly. The program is subject to internal peer review and can be evaluated by the user online.nnnCONCLUSIONnThe authors experience shows that a highly interactive Web-based training program for radiology, tailored to the requirements of the target group, can be developed economically by a team of medical students using an advanced storing system, with the guidance of a radiologist and without the help of professionally trained computer experts.

Which continuous US scanning mode is optimal for the detection of vascularity in liver lesions when enhanced with a second generation contrast agent

OBJECTIVESnMicrobubble echo-enhancers help in the assessment of focal liver masses by enhancing the signal from blood vessels. A variety of linear and nonlinear scanning modes are now available, but it is unclear which is optimal. A controlled comparison was performed during the infusion of such an agent (SonoVue: Bracco, Milan, Italy).nnnMETHODS AND MATERIALSnTen patients with known focal liver lesions were studied. The diagnoses, confirmed on dual phase helical computed tomography (CT) at the same attendance were metastasis (n = 7), haemangioma (n = 2) and focal nodular hyperplasia FNH (n = 1). A dose of 12 ml SonoVue concentrated at 5 mg/ml was infused intravenously at a rate of 1 ml/min. The enhancement level was monitored with a continuous wave (CW) Doppler probe over the right radial artery and the intensity of the signal was registered at 1 s intervals. When a plateau of enhancement was reached, a single lesion in each patient was imaged using five different continuous scanning modes, fundamental grey scale (FGS); fundamental colour Doppler (FCD); fundamental power Doppler (FPD); second harmonic grey scale (HGS); and pulse inversion mode (Pim) using an HDI5000 scanner and C5-2 probe (ATL, Bothell, WA). The order of scanning modes was varied between patients using a predefined randomisation protocol. The videos (super video home system (SVHS)) were analysed offsite by two blinded readers, both experienced in contrast ultrasound of the liver. The readers were asked to score each mode in terms of its ability to detect vessels within/around the lesion at optimal enhancement. This was done using a ranking system (1, worst; 5, best) for each patient.nnnRESULTSnBoth observers scored FPD as the optimal imaging method, followed by Pim. (Scores summed across all patients, observer 1: FPD 48, Pim 42, FCD 37, HGS 21, FGS 10; observer 2: FPD 49, Pim 40, FCD 38, HGS 21, FGS 10). The differences from FPD were significant for FCD, HGS and FGS using a unpaired analysis of variance (ANOVA) comparison, with Bonferroni multiple corrections, (P<0.01, both observers). The differences between FPD and Pim were also significant both for observer 2 and for both observers combined (P<0.01), but did not reach significance for observer 1 (P = 0.19).nnnCONCLUSIONSnIn this study, FPD performed best, and the non-linear modes, performed continuously (pulse inversion and second HGS), showed no clear advantage.

Multiclassifier fusion in human brain MR segmentation: modelling convergence

Segmentations of MR images of the human brain can be generated by propagating an existing atlas label volume to the target image. By fusing multiple propagated label volumes, the segmentation can be improved. We developed a model that predicts the improvement of labelling accuracy and precision based on the number of segmentations used as input. Using a cross-validation study on brain image data as well as numerical simulations, we verified the model. Fit parameters of this model are potential indicators of the quality of a given label propagation method or the consistency of the input segmentations used.

Analysis of serial MR images of joints

Medical research and drug discovery rely increasingly on making comparisons between MR images from large numbers of subjects, often with multiple time points for each subject. We show how image registration and visualisation technique can be used for quantitative and qualitative assessment of changes over time in an experimental model of rheumatoid arthritis. We applied the technique to automatically delineate two separate bones (the calcaneus and talus) from the in vivo MR images of an ankle (6 subjects, 6 time points each) by propagating labels from pre-labelled atlas MRI scans. As an initial analysis, we calculated their volume and rendered their surfaces in 3D to study the disease progression over time. Finally, we investigated the use of grid computing by coupling it with image analysis and visualisation algorithms to enable remote invocation, parallel execution, and source data input from heterogeneous distributed image repositories.

Enhancement characteristics of the microbubble agent Levovist: reproducibility and interaction with aspirin

We investigated the reproducibility of Doppler enhancement indices following intravenous bolus injections of Levovist (Schering AG, Berlin) microbubbles. We also aimed to determine whether observations from animal studies suggesting that aspirin potentiates microbubble enhancement were reproducible in humans. In five healthy volunteers, time enhancement profiles of Doppler intensity following repeated bolus injections of Levovist were acquired from the common carotid artery, hepatic vein and kidney using spectral and power Doppler before and after oral aspirin (600 mg). Peak enhancement (PE), area under the curve (AUC) and decay slopes (lambda) were calculated. Hepatic vein contrast arrival time (AT) was determined subjectively. Well-defined carotid enhancement was seen in 19/20 injections. Reproducibility was high (r > 0.8). PE and AUG were unaffected by aspirin, but lambda was slightly reduced (P = 0.02). Renal power Doppler profiles were well defined (10/10) with no significant changes of AUC, PE or lambda after aspirin. Our study demonstrates good reproducibility of carotid spectral Doppler time intensitometry with Levovist in man. Aspirin does not have a significant effect on enhancement indices except carotid spectral Doppler decay. We conclude that aspirin is unlikely to potentiate microbubble enhancement, as seen in animal studies.

Propagating labels of the human brain based on non-rigid MR image registration: an evaluation

Background: In order to perform statistical analysis of cohorts based on images, reliable methods for automated anatomical segmentation are required. Label propagation (LP) from manually segmented atlases onto newly acquired images is a particularly promising approach. Methods: We investigated LP on a set of 6 three-dimensional T1-weighted magnetic resonance data sets of the brains of normal individuals. For each image, a manually prepared segmentation of 67 structures was available. Each subject image was used in turn as an atlas and registered non-rigidly to each other subjects image. The resulting transformations were applied to the label sets, yielding five different generated segmentations for each subject, which we compared with the native manual segmentations using an overlap measure (similarity index, SI). We then reviewed the LP results for five structures with varied anatomical and label characteristics visually to determine how the registration procedure had affected the delineation of their boundaries. Results: The majority of structures propagated well as measured by SI (SI > 70 in 80% of measurements). Boundaries that were marked in the atlas image by definite intensity differences were congruent, with good agreement between the manual and the generated segmentations. Some boundaries in the manual segmentation were defined as planes marked by landmarks; such boundaries showed greater mismatch. In some cases, the proximity of structures with similar intensity distorted the LP results: e.g., parts of the parahippocampal gyrus were labeled as hippocampus in two cases. Conclusion: The size and shape of anatomical structures can be determined reliably using label propagation, especially where boundaries are defined by distinct differences in grey scale image intensity. These results will inform further work to evaluate potential clinical uses of information extracted from images in this way.