Medical Visionday Abstracts

Return to the Medical Visionday.

Session I
09.10	NeatVision - Visual programming for computer aided diagnostic applications Paul Whelan, Vision Systems Group, School of Engineering, Dublin City University, Ireland Keynote address This talk will detail a free visual programming based image analysis development environment for medical imaging applications. The environment provides high-level access to a wide range of image processing algorithms through a well-defined and easy to use graphical interface. The system contains over 300 image manipulation, processing and analysis algorithms. For more advanced users an upgrade path is provided to extend the core library using the developer’s interface. This additional freely available plug-in features, automatic source code generation, compilation with full error feedback and dynamic algorithm updates. The talk will also discuss key issues associated with the environment and outline the advantages in adopting such a system for computer aided diagnostic application development. A wide range of computer aided diagnostic sample applications will also be presented to illustrate the flexibility of the environment. Extended abstract with figures (pdf)
10.00	On automating and standardising corpus callosum analysis in brain MRI Karl Skoglund, IMM, DTU Corpus callosum analysis is influenced by many factors. The effort in controlling these has previously been incomplete and scattered. This paper sketches a complete pipeline for automated corpus callosum analysis from magnetic resonance images, with focus on measurement standardisation. The presented pipeline deals with i) estimation of the mid-sagittal plane, ii) localisation and registration of the corpus callosum, iii) parameterisation and representation of its contour, and iv) means of standardising the traditional reference area measurements.
Session II
11.00	Estimation of complex motions in video signals Dr.Ing. Erhardt Barth, Institute for Neuro- and Bioinformatics, University of Lübeck, Germany Presentation
11.20	Automated histomorphometric analysis of joint damage in a mouse model of osteoarthritis Thomas Jensen, Visiopharm, Denmark Morphometric analysis of histological sections of the knee is essential for quantifying the degree of joint damage and drug efficacy in animal models of osteoarthritis (OA). It is of particular interest to measure the number of cells and the area of the cartilage lesions as they are manifested in the histological sections. Manual counting of cells is very time consuming, and scoring of cartilage loss is qualitative and subject to grader variability. The aim of the work presented here was to develop image analysis software capable of performing a fully automated segmentation of Hematoxilin/Eosin (H/E) stained histological knee sections, thus allowing for measuring various morphometric variables, including cellularity and lesion area. 220 light microscopy images were of H/E stained histological knee sections obtained from 50 male STR/1N mice, a strain which develops OA spontaneously, were used for developing the software algorithms. In order to develop a method which is robust towards variation in staining intensity, knowledge about the anatomy of the knee, was incorporated into the algorithm. This allows for an automated localization of regions for training the chosen pixel classifier for recognizing bone matrix, cartilage, cells (and lacunae), and background. The automated training is repeated for each individual image. Position and shape of the cartilage areas are exploited to perform correct delineation, of the femoral and tibial cartilage. Objects that are irrelevant for morphometric measurements, such as e.g. the meniscus, are automatically excluded, by an iterative local curve fitting approach. Cells inside the cartilage areas are segmented using a special type of filters giving a strong response, wherever small rounded objects are present. An iterative method, locating local maxima allows for correct counting even when cells are partly overlapping. A novel image analysis method was developed, capable of performing an automated segmenting of H/E stained histological knee sections spatially into anatomically correct areas of bone, cartilage, cells (chondrocytes), lacunae, and background. Based on this segmentation, the system is capable of measuring cell number, cartilage area, surface length, and a number of other morphometric variables that can be defined by the scientists using the software, for developing and testing scientific hypotheses. The user interface for the system allows the scientists to inspect each image for correctness of the computerized segmentation as well as the measurements. It is simple and intuitive to perform corrections in those cases where corrections may be deemed necessary. The developed method is flexible enough to allow for different staining methods. If the promising first results of this automated image analysis method can be verified on a larger test set, a new and powerful research tool will be available to scientists for performing histomorphometric measurements much faster, and most likely with a higher degree of reproducibility.
11.40	Systemic vascular diseased assessed by noninvasive observation of the eye Michael Larsen, Dept. of Opthalmology, Herlev Hospital, Copenhagen
Session III: Sponsored by the Biophotonics Network
13.30	Improved OCT imaging with a new averaging method Birgit Sander, Dept. of Opthalmology, Herlev Hospital, Copenhagen Birgit Sander¹, Michael Larsen¹, Thomas Martini Jørgensen² ¹Dept. Of Opthalmology, Herlev Hospital, Copenhagen, ²Risø National Laboratory, Roskilde Purpose: To investigate the effect on image quality of a new averaging method for OCT3 scans. Methods: A dominant source of noice in OCT is the signal from nearby scatteres in the measuring volume of the OCT scan. Repeated OCT-scans (Stratus, Carl Zeiss, Meditec) were obtained with 512 points per 6mm line scan in the line-mode, approximately 15 times. An averaging algoritm was applied to reduce noise (speckle) with a robust correlation algoritm using a maximising a least square measure of the diffence between corresponding A-scans. For detailed examination the images were shown in gray and color scale. The method was evalated on healthy subjects, patients with retinal edema due to diabetes, macular hole centrally in the retina and central serous chorioretinopathy. Results: The averaged images showed a distinctive appearance and separation of the retinal layers in comparison to standard images, ie. the high reflectance of retinal synaptic layers (the plexifom layers, the external limiting membrane and the junction between inner and outer photoreceptor segments) was improved in contrast against the low-electing nuclear layers. Also, the outer photoreceptor and retinal pigment epithelium (RPE) layers showed more details than the standard image and 3 hyperreflective and two hyporeflective layers are now discernible. From cases of healthy subjects and chorideal serous retinopathy the 3 hyperreflective layers of the outer OCT band can now be attributed to the junction of inner/outer photoreceptor segments, the interdigitising outer photoreceptor segments / RPE villi and the outermost layer the cell bodies of the RPE. Whether or not a signal from the choriocapillaris is included in this outermost layer has yet to be determinded. Conclusion: Averaging of OCT3/stratus scans improves the visual quality considerably and allows a detailed interpretation of the retinal layers superior to standard scans allowing a detailed evaluation of the location of pathological changes. Averaged images are in particular useful to study whether the outer parts of photoreceptor layers or defects of the retinal pigment epithelium is present in the individual patient.
13.50	Active contour model for segmenting OCT B-scans Jakob Thomadsen, Risø National Laboratory Presentation
14.10	High-resolution, real-time imaging using a fiber-based optical coherence tomography system Peter Falk, Research Center COM, Technical University of Denmark Peter Falk¹, Michael Frosz¹, Lars Thrane², Ole Bang1, Peter E. Andersen², and Anders Bjarklev² ¹Research Center COM, Technical University of Denmark, DK–2800 Kgs. Lyngby ²Optics and Plasma Research Department, Risø National Laboratory, DK–4000 Roskilde, Denmark A fiber based optical coherence tomography system has been developed for the purpose of biomedical imaging of skin. The system is targeted to be a compact, modular system and produces real-time images with 8 frames/second. The deliberate features give a state-of-the-art imaging system and advantages for experimental research on the development of system modules, e.g. supercontinuum light sources. We will present our initial results from the system and its usability for light source characterization.
14.30	Biomedical sensors based on optical photonic crystal fibres Lars Rindorf, Research Center COM, Technical University of Denmark Photonic crystal fibres have small internal holes along the axis. These holes can be used as cappilary tubes and their contents can be probed by the optical properties of the fibre. At COM we have used this technique for biomolecule sensing, where the cappilary tubes are coated with a DNA-speci_c receptor coating. The system has been integrated by the group in a versatile lab-on-a-chip unit, which allows for continous measurement of the fibre's contents of biomolecules. For more complex biosensoring, optical photonic crystal fiber based filters are under development.
14.50	Medical Image Computing and Computer-Assisted Intervention - MICCAI 2006, Copenhagen Rasmus Larsen, IMM, DTU MICCAI is the premier international conference with in-depth papers on medical image computing, computer-assisted intervention and medical robotics. The multidisciplinary nature of these emerging fields brings together clinicians, bioscientists, computer scientists, engineers, physicists & other researchers who are contributing to, and need to keep abreast of, advances in the methodology and applications. (more)

Session IV
15.30	The virtual slaughterhouse – registration and segmentation of 3D CT scans of pig bodies Mads Fogtmann Hansen, IMM, DTU Quality estimation of meat today is an expensive, tedious and labour full task. Moreover, the methods used today are mostly applied after the final cutting. Hence, it is not possible to use these estimation methods as a decision base for choosing the most profitable way of cutting out the meat. The significant advantages within non-invasive visualization techniques in the last decade, such as CT scanning and MR imaging, open new doors in the aim of developing a reliable and automatic method for quality estimation of meat prior to the final cutting. Obviously, a good and possible segmentation into meat, fat and bone is needed before the actual quality estimation can be carried out. This talk will focus on segmentation of the back of the pig which is the most valuable part. One approach is to use a statistical shape model to guide and constrain the segmentation, enabling us to produce a likely segmentation in areas lacking information. One key consideration in building a statistical shape model is to ensure that the model captures all significant variance. Hence, if the training set is not representative the model will be over constrained. This is a major problem with slaughtered pig backs as areas away from the ribs are deformed by gravity. It is difficult, perhaps even impossible, to capture this non-anatomical variance in a shape model, why it is necessary to remove the non-anatomical variance before a shape model guided segmentation can be applied. A novel method for removal of non-anatomical variance will be presented. Likewise, a method for finding and removing ribs from the CT scan of the back will be presented. The reason for removing the ribs is that the ribs have no influence on the quality of the meat, however they might have unwanted effect on the quality estimation scheme.
15.50	Porcine obesity – fat segmentation in multimodality 3D MR abdominal MR scans Rasmus Engholm, IMM, DTU Obesity is by WHO considered one of the most rapidly spreading non-infectious epidemics world-wide. Obesity and especially the accumulation of intra-abdominal fat leads to a number of metabolic disturbances in humans, best known is type 2 diabetes, which is estimated to affect close to 300 millions by 2025. A study of mini pigs carried out at KVL is trying to determine the effect of different diets on the different fat types. In order to validate the effect over time an in vivo method of assessing the amount of different fat types is needed. The pigs have been scanned at an MR-scanner at Hvidovre Hospital, producing two different types of MR images (T1 and T1 HASTE IRM). The goal is to try to asses the different fat types based on these images, and compare our result with the manually dissected pigs. The talk will cover some aspects of preprocessing the MR-images including bias correction and coregistration and some segmentation of fat.

Return to the Medical Visionday.

Visionday frontpage | Further information: Secretary Eina Boeck, info@visionday.dk, Ph. +45 4535 3424