Jens N. Kaftan

Automatic multi-organ segmentation using learning-based segmentation and level set optimization

We present a novel generic segmentation system for the fully automatic multi-organ segmentation from CT medical images. Thereby we combine the advantages of learning-based approaches on point cloud-based shape representation, such a speed, robustness, point correspondences, with those of PDE-optimization-based level set approaches, such as high accuracy and the straightforward prevention of segment overlaps. In a benchmark on 10-100 annotated datasets for the liver, the lungs, and the kidneys we show that the proposed system yields segmentation accuracies of 1.17-2.89 mm average surface errors. Thereby the level set segmentation (which is initialized by the learning-based segmentations) contributes with an 20%-40% increase in accuracy.

Fuzzy pulmonary vessel segmentation in contrast enhanced CT data

Pulmonary vascular tree segmentation has numerous applications in medical imaging and computer-aided diagnosis (CAD), including detection and visualization of pulmonary emboli (PE), improved lung nodule detection, and quantitative vessel analysis. We present a novel approach to pulmonary vessel segmentation based on a fuzzy segmentation concept, combining the strengths of both threshold and seed point based methods. The lungs of the original image are first segmented and a threshold-based approach identifies core vessel components with a high specificity. These components are then used to automatically identify reliable seed points for a fuzzy seed point based segmentation method, namely fuzzy connectedness. The output of the method consists of the probability of each voxel belonging to the vascular tree. Hence, our method provides the possibility to adjust the sensitivity/specificity of the segmentation result a posteriori according to application-specific requirements, through definition of a minimum vessel-probability required to classify a voxel as belonging to the vascular tree. The method has been evaluated on contrast-enhanced thoracic CT scans from clinical PE cases and demonstrates overall promising results. For quantitative validation we compare the segmentation results to randomly selected, semi-automatically segmented sub-volumes and present the resulting receiver operating characteristic (ROC) curves. Although we focus on contrast enhanced chest CT data, the method can be generalized to other regions of the body as well as to different imaging modalities.

A novel multipurpose tree and path matching algorithm with application to airway trees

Tree matching methods have numerous applications in medical imaging, including registration, anatomical labeling, segmentation, and navigation of structures such as vessels and airway trees. Typical methods for tree matching rely on conventional graph matching techniques and therefore suffer potential limitations such as sensitivity to the accuracy of the extracted tree structures, as well as dependence on the initial alignment. We present a novel path-based tree matching framework independent of graph matching. It is based on a point-by-point feature comparison of complete paths rather than branch points, and consequently is relatively unaffected by spurious airways and/or missing branches. A matching matrix is used to enforce one-to-one matching. Moreover our method can reliably match irregular tree structures, resulting from imperfect segmentation and centerline extraction. Also reflecting the nature of these features, our method does not require a precise alignment or registration of tree structures. To test our method we used two thoracic CT scans from each of ten patients, with a median inter-scan interval of 3 months (range 0.5 to 10 months). The bronchial tree structure was automatically extracted from each scan and a ground truth of matching paths was established between each pair of tree structures. Overall 87% of 702 airway paths (average 35.1 per patient matched both ways) were correctly matched using this technique. Based on this success we also present preliminary results of airway-to-artery matching using our proposed methodology.

Multi-stage learning for robust lung segmentation in challenging CT volumes

Simple algorithms for segmenting healthy lung parenchyma in CT are unable to deal with high density tissue common in pulmonary diseases. To overcome this problem, we propose a multi-stage learning-based approach that combines anatomical information to predict an initialization of a statistical shape model of the lungs. The initialization first detects the carina of the trachea, and uses this to detect a set of automatically selected stable landmarks on regions near the lung (e.g., ribs, spine). These landmarks are used to align the shape model, which is then refined through boundary detection to obtain fine-grained segmentation. Robustness is obtained through hierarchical use of discriminative classifiers that are trained on a range of manually annotated data of diseased and healthy lungs. We demonstrate fast detection (35s per volume on average) and segmentation of 2 mm accuracy on challenging data.

A Two-Stage Approach for Fully Automatic Segmentation of Venous Vascular Structures in Liver CT Images

The segmentation of the hepatic vascular tree in computed tomography (CT) images is important for many applications such as surgical planning of oncological resections and living liver donations. In surgical planning, vessel segmentation is often used as basis to support the surgeon in the decision about the location of the cut to be performed and the extent of the liver to be removed, respectively. We present a novel approach to hepatic vessel segmentation that can be divided into two stages. First, we detect and delineate the core vessel components efficiently with a high specificity. Second, smaller vessel branches are segmented by a robust vessel tracking technique based on a medialness filter response, which starts from the terminal points of the previously segmented vessels. Specifically, in the first phase major vessels are segmented using the globally optimal graphcuts algorithm in combination with foreground and background seed detection, while the computationally more demanding tracking approach needs to be applied only locally in areas of smaller vessels within the second stage. The method has been evaluated on contrast-enhanced liver CT scans from clinical routine showing promising results. In addition to the fully-automatic instance of this method, the vessel tracking technique can also be used to easily add missing branches/sub-trees to an already existing segmentation result by adding single seed-points.

High Dynamic Range Microscopy for Cytopathological Cancer Diagnosis

Cancer is one of the most common causes of death. Cytopathological, i.e., cell-based, diagnosis of cancer can be applied in screening scenarios and allows an early and highly sensitive detection of cancer, thus increasing the chance for cure. The detection of cancer on cells addressed in this paper is based on bright field light microscopy. The cells are imaged with a camera mounted on a microscope, allowing to measure cell properties. However, these cameras exhibit only a limited dynamic range, which often makes the quantification of properties difficult or even impossible. Consequently, to allow a computer-assisted analysis of microscopy images, the imaging has to be improved. To this end, we show how the dynamic range can be increased by acquiring a set of differently exposed cell images. These high dynamic range (HDR) images allow to measure cellular features that are otherwise difficult to capture, if at all. We show that HDR microscopy not only increases the dynamic range, but furthermore reduces noise and improves the acquisition of colors. We develop HDR microscopy-based algorithms, which are essential for cytopathological oncology and early cancer detection and only possible with HDR microscopy imaging. We show the detection of certain subcellular features, so-called AgNORs, in silver (Ag) stained specimens. Furthermore, we give examples of two further applications, namely: 1) the detection of stained cells in immunocytochemical preparations and 2) color separation for nuclear segmentation of specimens stained with low contrast.

Noise in high dynamic range imaging

High dynamic range (HDR) imaging is more and more widely used to increase the limited dynamic range of digital cameras and, in turn, to cover the dynamic range of the acquired scene. This image acquisition process can be subdivided into two steps. The first step is the measurement or estimation of the mostly non-linear camera transfer function (CTF). This is followed by the second step, the combination of a set of differently exposed images of the same scene into one HDR image after linearization with the inverse CTF. Each of the individual images in such an exposure set contains noise from the image acquisition process. Consequently, the calculated HDR image will as well contain noise, which fortunately is reduced by the weighted average of the images from the exposure set. We analyze the achieved gain in SNR for different weighting functions proposed in the literature and compare these with a plain average. Although these functions are based on reasonable intuitions, we show that the highest SNRgain is achieved with the plain average.

An Evaluation Framework for the Accuracy of Camera Transfer Functions Estimated from Differently Exposed Images

Intensity values read from CCD- or CMOS-cameras are usually not proportional to the irradiance acquired by the sensor, but are mapped by a mostly nonlinear camera transfer function (CTF). This CTF can be measured using a test chart. This, however, is afflicted with the difficulty of ensuring uniform illumination of the chart. An alternative to chart-based measurements of the CTF is to use differently exposed images of the same scene. In this paper, we describe a radiometry-based experimental setup to directly measure the CTF. We furthermore show how to obtain image pairs of known exposure ratios from the same experiment, i.e., under identical environmental circumstances (light, temperature, camera settings). We use these images to estimate the CTF on differently exposed images, thus enabling a quantitative comparison of estimated and measured CTF

Fast self-collision detection and simulation of bifurcated stents to treat abdominal aortic aneurysms (AAA)

In this paper, we present a fast and robust collision detection algorithm for bifurcated stent graft simulation. The algorithm is designed for simulating bifurcations of Y-stents in blood vessels in CT-data. For stent simulation, we use the active contours method (ACM) which realizes internal and external forces acting on the stent. The internal forces are for the elastic and smoothing behavior of the stent surface. The external forces (balloon force and resistance of the vessel wall) pull and push the stent towards the artery wall. In addition to these two forces, another force interacts in the bifurcated part of the stent. For this purpose, the algorithm uses axis aligned bounding boxes (AABB) that are placed along these bifurcated parts. Our simulation and collision detection algorithm is evaluated on computed tomography angiography (CTA) data with variations in anatomy and location of pathology.

High Dynamic Range Images as a Basis for Detection of Argyrophilic Nucleolar Organizer Regions Under Varying Stain Intensities

Silver staining of cytopathologic specimens offers advantages in cancer diagnostics. A difficulty with such stained cell specimens is the very high dynamic range needed by the imaging system to appropriately cover the varying stain intensities. Beside those images of cell nuclei that can be used for the diagnostic interpretation, there are nuclei that appear too dark to observe their relevant properties, the so-called argyrophilic nucleolar organizer regions (AgNORs), which appear as spot-like areas darker than their immediate surroundings. We therefore show how high dynamic range images of nuclei can help to correctly segment the AgNORs. To this end, we acquire a sequence of differently exposed images, which are then combined into a high dynamic range image. Based on the dynamic range of the image signal within the segmented cell area, we compute another image which provides optimal contrast over this area of interest. To further increase the contrast for dark objects, a suitable nonlinear point transform is simultaneously applied. We provide examples of the thus generated images and their corresponding segmentations.