Bjarne Kjær Ersbøll

Quantification of biofilm structures by the novel computer program COMSTAT

The structural organization of four microbial communities was analysed by a novel computer program, COMSTAT, which comprises ten features for quantifying three-dimensional biofilm image stacks. Monospecies biofilms of each of the four bacteria, Pseudomonas: putida, P. aureofaciens, P. fluorescens and P. aeruginosa, tagged with the green fluorescent protein (GFP) were grown in flow chambers with a defined minimal medium as substrate. Analysis by the COMSTAT program of four variables describing biofilm structure - mean thickness, roughness, substratum coverage and surface to volume ratio - showed that the four Pseudomonas: strains represent different modes of biofilm growth. P. putida had a unique developmental pattern starting with single cells on the substratum growing into micro-colonies, which were eventually succeeded by long filaments and elongated cell clusters. P. aeruginosa colonized the entire substratum, and formed flat, uniform biofilms. P. aureofaciens resembled P. aeruginosa, but had a stronger tendency to form micro-colonies. Finally, the biofilm structures of P. fluorescens had a phenotype intermediate between those of P. putida and P. aureofaciens. Analysis of biofilms of P. aureofaciens growing on 0.03 mM, 0.1 mM or 0.5 mM citrate minimal media showed that mean biofilm thickness increased with increasing citrate concentration. Moreover, biofilm roughness increased with lower citrate concentrations, whereas surface to volume ratio increased with higher citrate concentrations.

Sparse Discriminant Analysis

We consider the problem of performing interpretable classification in the high-dimensional setting, in which the number of features is very large and the number of observations is limited. This setting has been studied extensively in the chemometrics literature, and more recently has become commonplace in biological and medical applications. In this setting, a traditional approach involves performing feature selection before classification. We propose sparse discriminant analysis, a method for performing linear discriminant analysis with a sparseness criterion imposed such that classification and feature selection are performed simultaneously. Sparse discriminant analysis is based on the optimal scoring interpretation of linear discriminant analysis, and can be extended to perform sparse discrimination via mixtures of Gaussians if boundaries between classes are nonlinear or if subgroups are present within each class. Our proposal also provides low-dimensional views of the discriminative directions.

FAME-a flexible appearance modeling environment

Combined modeling of pixel intensities and shape has proven to be a very robust and widely applicable approach to interpret images. As such the active appearance model (AAM) framework has been applied to a wide variety of problems within medical image analysis. This paper summarizes AAM applications within medicine and describes a public domain implementation, namely the flexible appearance modeling environment (FAME). We give guidelines for the use of this research platform, and show that the optimization techniques used renders it applicable to interactive medical applications. To increase performance and make models generalize better, we apply parallel analysis to obtain automatic and objective model truncation. Further, two different AAM training methods are compared along with a reference case study carried out on cross-sectional short-axis cardiac magnetic resonance images and face images. Source code and annotated data sets needed to reproduce the results are put in the public domain for further investigation.

Experimental reproducibility in flow-chamber biofilms

The structural organization of microbial communities is influenced by many factors, e.g. nutrient composition, shear stress and temperature. This paper presents a general method for quantitative comparison of biofilm structures and assessment of experimental reproducibility between independent biofilm experiments. By using a novel computer program, COMSTAT, biofilm structures of Pseudomonas aeruginosa and an isogenic rpoS mutant were quantified. The strains were tagged with the green fluorescent protein (GFP) and grown in flow chambers with a defined minimal medium as substrate. Three independent rounds of biofilm experiments were performed and in each round, each of the two variants was grown in two separate channels. Nine image stacks were acquired in each channel 146 h after inoculation. An analysis of variance model incorporating the factors experiment round, bacterial strain, channel number and image stack number was used to analyse the data calculated by COMSTAT. Experimental reproducibility was verified by estimating the magnitude of the variance of the effects round (sigma(2)R) and the interaction between bacterial strain and round (sigma(2)BR). Mean thickness of the wild-type and rpoS mutant biofilms was estimated at 6.31 microm (SE 0.81 microm) and 16.85 microm (SE 0.87 microm), respectively.

Surface-bounded growth modeling applied to human mandibles

From a set of longitudinal three-dimensional scans of the same anatomical structure, the authors have accurately modeled the temporal shape and size changes using a linear shape model. On a total of 31 computed tomography scans of the mandible from six patients, 14,851 semilandmarks are found automatically using shape features and a new algorithm called geometry-constrained diffusion. The semilandmarks are mapped into Procrustes space. Principal component analysis extracts a one-dimensional subspace, which is used to construct a linear growth model. The worst case mean modeling error in a cross validation study is 3.7 mm.

Independent Histogram Pursuit for Segmentation of Skin Lesions

In this paper, an unsupervised algorithm, called the Independent Histogram Pursuit (IHP), for segmenting dermatological lesions is proposed. The algorithm estimates a set of linear combinations of image bands that enhance different structures embedded in the image. In particular, the first estimated combination enhances the contrast of the lesion to facilitate its segmentation. Given an N-band image, this first combination corresponds to a line in N dimensions, such that the separation between the two main modes of the histogram obtained by projecting the pixels onto this line, is maximized. The remaining combinations are estimated in a similar way under the constraint of being orthogonal to those already computed. The performance of the algorithm is tested on five different dermatological datasets. The results obtained on these datasets indicate the robustness of the algorithm and its suitability to deal with different types of dermatological lesions. The boundary detection precision using k-means segmentation was close to 97%. The proposed algorithm can be easily combined with the majority of classification algorithms.

Quantitative measurement of changes in retinal vessel diameter in ocular fundus images

The change in diameter of retinal vessels as a function of increasing distance to the optic disc is believed to be indicative of the risk level of various vascular diseases such as generalised arteriosclerosis and Diabetes Mellitus. In particular, focal arteriolar narrowing (FAN) is considered related to arteriosclerosis. The aim of this work is to develop methods to provide quantitative information about the FAN status of retinal arteriolar vessel segments. We propose a method to measure the vessel diameter and changes herein along the vessel. The width or diameter measurement is based on intensity profiles orthogonal to a smoothed trace in the vessel found by Dijkstra’s shortest path algorithm. The vessel diameter is calculated from the intensity profiles using two diAerent methods to estimate profile widths. We propose a ‘‘normalised accumulated gradient’’ (NAG) and the coeAcient of variance (CV) to estimate the FAN in a vessel segment. The NAG is designed to detect increases in the vessel diameter as the distance to the papilla increases. The performance of the methods developed is compared to the evaluation by a skilled ophthalmologist. The methods are seen to perform well. ” 2000 Elsevier Science B.V. All rights reserved.

Multispectral Imaging for Determination of Astaxanthin Concentration in Salmonids

Multispectral imaging has been evaluated for characterization of the concentration of a specific cartenoid pigment; astaxanthin. 59 fillets of rainbow trout, Oncorhynchus mykiss, were filleted and imaged using a rapid multispectral imaging device for quantitative analysis. The multispectral imaging device captures reflection properties in 19 distinct wavelength bands, prior to determination of the true concentration of astaxanthin. The samples ranged from 0.20 to 4.34 g per g fish. A PLSR model was calibrated to predict astaxanthin concentration from novel images, and showed good results with a RMSEP of 0.27. For comparison a similar model were built for normal color images, which yielded a RMSEP of 0.45. The acquisition speed of the multispectral imaging system and the accuracy of the PLSR model obtained suggest this method as a promising technique for rapid in-line estimation of astaxanthin concentration in rainbow trout fillets.

Virtual dissection of pig carcasses

This paper proposes the use of computed tomography (CT) as a reference method for estimating the lean meat percentage (LMP) of pig carcasses. The current reference is manual dissection which has a limited accuracy due to variability between butchers. A contextual Bayesian classification scheme is applied to classify volume elements of full body CT-scans of pig carcasses into three tissue types. A linear model describes the relation between voxels and the full weight of the half carcass, which can be determined more accurately than that of the lean meat content. Two hundred and ninety-nine half pig carcasses were weighed and CT-scanned. The explained variance of the model was R(2)=0.9994 with a root-mean-squared error of prediction of 83.6g. Applying this method as a reference will ensure a more robust calibration of sensors for measuring the LMP, which is less prone to variation induced by manual intervention.

Building and Testing a Statistical Shape Model of the Human Ear Canal

Today the design of custom in-the-ear hearing aids is based on personal experience and skills and not on a systematic description of the variation of the shape of the ear canal. In this paper it is described how a dense surface point distribution model of the human ear canal is built based on a training set of laser scanned ear impressions and a sparse set of anatomical landmarks placed by an expert. The landmarks are used to warp a template mesh onto all shapes in the training set. Using the vertices from the warped meshes, a 3D point distribution model is made. The model is used for testing for gender related differences in size and shape of the ear canal.