Bettina Selig

Improving the stochastic watershed

The stochastic watershed is an unsupervised segmentation tool recently proposed by Angulo and Jeulin. By repeated application of the seeded watershed with randomly placed markers, a probability density function for object boundaries is created. In a second step, the algorithm then generates a meaningful segmentation of the image using this probability density function. The method performs best when the image contains regions of similar size, since it tends to break up larger regions and merge smaller ones. We propose two simple modifications that greatly improve the properties of the stochastic watershed: (1) add noise to the input image at every iteration, and (2) distribute the markers using a randomly placed grid. The noise strength is a new parameter to be set, but the output of the algorithm is not very sensitive to this value. In return, the output becomes less sensitive to the two parameters of the standard algorithm. The improved algorithm does not break up larger regions, effectively making the algorithm useful for a larger class of segmentation problems.

Fully automatic evaluation of the corneal endothelium from in vivo confocal microscopy

BackgroundManual and semi-automatic analyses of images, acquired in vivo by confocal microscopy, are often used to determine the quality of corneal endothelium in the human eye. These procedures are highly time consuming. Here, we present two fully automatic methods to analyze and quantify corneal endothelium imaged by in vivo white light slit-scanning confocal microscopy.MethodsIn the first approach, endothelial cell density is estimated with the help of spatial frequency analysis. We evaluate published methods, and propose a new, parameter-free method. In the second approach, based on the stochastic watershed, cells are automatically segmented and the result is used to estimate cell density, polymegathism (cell size variability) and pleomorphism (cell shape variation). We show how to determine optimal values for the three parameters of this algorithm, and compare its results to a semi-automatic delineation by a trained observer.ResultsThe frequency analysis method proposed here is more precise than any published method. The segmentation method outperforms the fully automatic method in the NAVIS software (Nidek Technologies Srl, Padova, Italy), which significantly overestimates the number of cells for cell densities below approximately 1200 mm−2, as well as previously published methods.ConclusionsThe methods presented here provide a significant improvement over the state of the art, and make in vivo, automated assessment of corneal endothelium more accessible. The segmentation method proposed paves the way to many possible new morphometric parameters, which can quickly and precisely be determined from the segmented image.

Influence of applied corneal endothelium image segmentation techniques on the clinical parameters

The corneal endothelium state is verified on the basis of an in vivo specular microscope image from which the shape and density of cells are exploited for data description. Due to the relatively low image quality resulting from a high magnification of the living, non-stained tissue, both manual and automatic analysis of the data is a challenging task. Although, many automatic or semi-automatic solutions have already been introduced, all of them are prone to inaccuracy. This work presents a comparison of four methods (fully-automated or semi-automated) for endothelial cell segmentation, all of which represent a different approach to cell segmentation; fast robust stochastic watershed (FRSW), KH method, active contours solution (SNAKE), and TOPCON ImageNET. Moreover, an improvement framework is introduced which aims to unify precise cell border location in images pre-processed with differing techniques. Finally, the influence of the selected methods on clinical parameters is examined, both with and without the improvement framework application. The experiments revealed that although the image segmentation approaches differ, the measures calculated for clinical parameters are in high accordance when CV (coefficient of variation), and CVSL (coefficient of variation of cell sides length) are considered. Higher variation was noticed for the H (hexagonality) metric. Utilisation of the improvement framework assured better repeatability of precise endothelial cell border location between the methods while diminishing the dispersion of clinical parameter values calculated for such images. Finally, it was proven statistically that the image processing method applied for endothelial cell analysis does not influence the ability to differentiate between the images using medical parameters.

Exact evaluation of stochastic watersheds : From trees to general graphs

The stochastic watershed is a method for identifying salient contours in an image, with applications to image segmentation. The method computes a probability density function (PDF), assigning to each piece of contour in the image the probability to appear as a segmentation boundary in seeded watershed segmentation with randomly selected seedpoints. Contours that appear with high probability are assumed to be more important. This paper concerns an efficient method for computing the stochastic watershed PDF exactly, without performing any actual seeded watershed computations. A method for exact evaluation of stochastic watersheds was proposed by Meyer and Stawiaski (2010). Their method does not operate directly on the image, but on a compact tree representation where each edge in the tree corresponds to a watershed partition of the image elements. The output of the exact evaluation algorithm is thus a PDF defined over the edges of the tree. While the compact tree representation is useful in its own right, it is in many cases desirable to convert the results from this abstract representation back to the image, e.g, for further processing. Here, we present an efficient linear time algorithm for performing this conversion.

Automatic measurement of compression wood cell attributes in fluorescence microscopy images

This paper presents a new automated method for analyzing compression wood fibers in fluorescence microscopy. Abnormal wood known as compression wood is present in almost every softwood tree harvested. Compression wood fibers show a different cell wall morphology and chemistry compared to normal wood fibers, and their mechanical and physical characteristics are considered detrimental for both construction wood and pulp and paper purposes. Currently there is the need for improved methodologies for characterization of lignin distribution in wood cell walls, such as from compression wood fibers, that will allow for a better understanding of fiber mechanical properties. Traditionally, analysis of fluorescence microscopy images of fiber cross‐sections has been done manually, which is time consuming and subjective. Here, we present an automatic method, using digital image analysis, that detects and delineates softwood fibers in fluorescence microscopy images, dividing them into cell lumen, normal and highly lignified areas. It also quantifies the different areas, as well as measures cell wall thickness. The method is evaluated by comparing the automatic with a manual delineation. While the boundaries between the various fiber wall regions are detected using the automatic method with precision similar to inter and intra expert variability, the position of the boundary between lumen and the cell wall has a systematic shift that can be corrected. Our method allows for transverse structural characterization of compression wood fibers, which may allow for improved understanding of the micro‐mechanical modeling of wood and pulp fibers.

Fast Evaluation of the Robust Stochastic Watershed

The stochastic watershed is a segmentation algorithm that estimates the importance of each boundary by repeatedly segmenting the image using a watershed with randomly placed seeds. Recently, this algorithm was further developed in two directions: (1) The exact evaluation algorithm efficiently produces the result of the stochastic watershed with an infinite number of repetitions. This algorithm computes the probability for each boundary to be found by a watershed with random seeds, making the result deterministic and much faster. (2) The robust stochastic watershed improves the usefulness of the segmentation result by avoiding false edges in large regions of uniform intensity. This algorithm simply adds noise to the input image for each repetition of the watershed with random seeds. In this paper, we combine these two algorithms into a method that produces a segmentation result comparable to the robust stochastic watershed, with a considerably reduced computation time. We propose to run the exact evaluation algorithm three times, with uniform noise added to the input image, to produce three different estimates of probabilities for the edges. We combine these three estimates with the geometric mean. In a relatively simple segmentation problem, F-measures averaged over the results on 46 images were identical to those of the robust stochastic watershed, but the computation times were an order of magnitude shorter.

Segmentation of Highly Lignified Zones in Wood Fiber Cross-Sections

Lignification of wood fibers has important consequences to the paper production, but its exact effects are not well understood. To correlate exact levels of lignin in wood fibers to their mechanical properties, lignin autofluorescence is imaged in wood fiber cross-sections. Highly lignified areas can be detected and related to the area of the whole cell wall. Presently these measurements are performed manually, which is tedious and expensive. In this paper a method is proposed to estimate the degree of lignification automatically. A multi-stage snake-based segmentation is applied on each cell separately. To make a preliminary evaluation we used an image which contained 17 complete cell cross-sections. This image was segmented both automatically and manually by an expert. There was a highly significant correlation between the two methods, although a systematic difference indicates a disagreement in the definition of the edges between the expert and the algorithm.