Fast Segmentation of Kidneys in CT Images

Size: px
Start display at page:

Download "Fast Segmentation of Kidneys in CT Images"

Transcription

1 WDS'10 Proceedings of Contributed Papers, Part I, 70 75, ISBN MATFYZPRESS Fast Segmentation of Kidneys in CT Images J. Kolomazník Charles University, Faculty of Mathematics and Physics, Prague, Czech Republic. Abstract. In this paper we present a fast implementation of segmentation algorithm suited for extraction of human kidneys in CT images. Kidneys have quite simple shape and that allows us to use faster algorithms than those which can adapt to changes in topology (e.g. level-sets). We based our 3D segmentation method on parametric snakes executed in volume slices to obtain contours, which define organs volume. This allows strong high-level parallelization of implemented method, which is scaling well with increasing number of processing units. We also decreased computational load by introducing more complex and descriptive statistical models, which substitute few expensive operations in segmentation process, reduce number of parameters and still give similar results. Introduction Precise segmentation of internal organs has various applications in medical practice visualization, diagnosis or medical treatment monitoring. We are going to present fast segmentation algorithm designed for segmentation of human kidneys from CT volume data. Segmentation of 3D objects from CT can be done in two ways. We can implement algorithm, which is three dimensional in its nature levelsets ((Adalsteinsson and Sethian, 1994)), subdivision surfaces ((Orderud and Rabben, 2008), 3D split/merge algorithms, watershed segmentation ((Straka et al., 2003)), etc. Or we can decide to use two dimensional segmentation algorithm executed in volume slices and coordinate these computations in some way. We chose second option for several reasons. 2D segmentation algorithms have longer tradition ((McInerney and Terzopoulos, 1996)) and are used more often, interaction with user is easier in two dimensions and our data have lower resolution in one dimension (CT images with thick slices). In the first section we are going to describe segmentation algorithm formally and in the second section we will talk about statistical models, which can be used for implementation of derived terms. 2D Segmentation Algorithm B-Spline Parametric Snakes As a base we chose parametric snakes from (Jacob et al., 2004), implemented in (Krajíček, 2007). Their version proved to be quite successful in precise segmentation but it was very slow and was controlled by large number of parameters we try to improve that. We used B-spline curve as contour representation. And we formulate energy term which for each contour shape computes amount of potential energy. This energy term is designed in such way that state with minimal energy defines optimal contour position. Optimization of this system is done by the iterative algorithm described in the end of this section. Energy terms A classical scheme of system energy is this: E snake = E image + E internal + E constraint (1) In following sections we describe these energy terms in detail. E image = E region + E edge (2) 70

2 Region energy term. Most important part of the image energy is region energy term. Its value is computed from region statistics and recent research on deformable models focuses on this type of term. Statistical approach gives us robust tool against noise, which is omnipresent in all real data computation doesn t depend on unstable image features, such as edges or corners. Main advantage of region based term is its large basin of attraction, so balloon force or similar terms, which are needed in other frameworks, can be omitted. If we need fast implementation we must prevent computation of region integrals, which is used in formulations of all region terms in this section. Mathematical analysis gives us tool called Green s Theorem, which allows us to convert region integrals to integrals along region contour. How to apply Green s Theorem can be found in (Jacob et al., 2004). As our first version of region energy term we use formulation from (Krajíček, 2007) and (Jacob et al., 2004): E region = log ( ) P(f(s) s Rin ) ds (3) P(f(s) s R out ) This integral gains its minimal value when represents region with higher probability P(f(s) s R in ) then P(f(s) s R out ). But this region term uses only distributions of intensity inside and outside of the segmented regions and ignores any information about position in space. To solve the previously mentioned problem we modify this term by using different probability function we use four dimensional distribution (intensity and spatial coordinates). ( ) P(f(s) s Rin ) E region2 = log ds (4) P(f(s) s R out ) There is one thing we can do with this complex model. We will need some basic formulas from conditional probability. If events A and B are independent following expressions are valid. P(A B) P(A B) = P(B) P(A B) = P(A B) P(B) (5) If we suppose that P(f(s)) and P(s R in ), respective P(f(s)) and P(s R out ) are independent, we can use 5. P(f(s) s R in ) = P(f(s) s R in ) P(s R in ) (6) From that we obtain: E region2 P(f(s) s R out ) = P(f(s) s R out ) P(s R out ) (7) ( ) ( ) P(f(s) s Rin ) P(s Rin ) = log + log ds (8) P(f(s) s R out ) P(s R out ) We can change this sum to weighted sum by introducing parameter α, which allows us to control influence of each sub-term: ( P(f(s) s Rin ) E region2 = α log P(f(s) s R out ) +(1 α) log ( P(s Rin ) P(s R out ) ) + ) ds (9) Notice special cases α = 1, we get the original 3 term, and α = 0, we get term whose value is controlled only by shape statistical model. Assumption of independence isn t really right. We try to segment homogeneous region, so independence of P(f(s)) and P(s R in ) is more or less kept. But outside of the region is situation different. Tissue around segmented organ is dependent on position, because organs are mostly at the same place and densities are correlated. Used simplification proofed sufficient, when segmenting kidneys of people with higher level of interorgan fat. Mathematically speaking, error caused by independence assumption isn t that big. But if we want to develop more general method we have to use the original formula 4 and not its simplifications 8 and 9. Later we will show how we can use these terms for fast initialization of parametric snakes. 71

3 Edge energy term. Region energy term is good for rough shape estimation. But for exact localization we must use some term, which is using image features. These terms have only small basin of attraction term has effect only when curve is close to edge. One of the most popular versions of edge energy term is based on integral of the square of the gradient magnitude along the curve. M E edgemag = f(v(t)) 2 dt (10) 0 Previously defined edge term was using only gradient magnitude and its direction was ignored. This flaw can be solved by different edge energy term: E grad = = C C k ( f(r) dr) (11) f(r) (dr k) (12) This one inceases energy if angle between contour normal and image gradient is too big. But this edge term is quite sensitive to noise. Internal and constrain energy term. In classic active contours algorithm ((Kass et al., 1988)), internal energy term was defined to maintain smoothness of the contour. But we are using parametric curves for shape representation so we have smooth derivatives by definition. But curves still can develop high curvature segments or collapse two control points into one - we need to prevent that. Derivation of suitable internal energy term can be found in (Jacob et al., 2004). Constrain energy term serves as tool for incorporating additional conditions, such as user input. In (Krajíček, 2007) was this term used for preventing curve divergence by shape restriction, generated from contours in previous slices. We did not use constraint energy term in our implementation. Optimization We used modified gradient steepest descent algorithm for locating point of minimum energy. But any other algorithm, which uses gradient of optimized function can be used. How to compute gradient of energy function can be found in (Kolomaznk, 2009; Krajíček, 2007; Jacob et al., 2004). Most of such numerical algorithms can end up in local extreme. So we need to initialize contour pretty close to the final shape. To prevent this we now effectively use weights of energy sub-terms. We can initialize our algorithm by arbitrary shape and run few iterations with boosted weight of shape energy sub-term (α in 9 is set to 0). We can also use spline with low number of control points to make it faster. In successive iterations we steadily increase influence of other region energy terms. This brings contour near the optimal shape. In the final phase we boost influence of edge energy term, which force spline to follow object contour. In our implementation one point initialization is enough we use it as center of circle with area similar to the final shape and then we execute optimization algorithm. See figure Figure 1. Segmentation steps in slice: 1 Init point computed from user input; 2 Snake initialized by circle with center in init point, radius computed from information stored in statistical model; 3 First phase of segmentation, controlled primarily by statistical shape model; 4 Second phase of segmentation, controlled mainly by image data. 72

4 Statistical Model KOLOMAZNÍK: FAST SEGMENTATION OF KIDNEYS IN CT IMAGES We formally deduced equations which control our segmentation process. Now we must find a way to represent our model in program. It is complicated to represent the model explicitly and precisely so we need to approximate probabilities somehow. First we want probabilities used in term 3. If we compute estimation of distribution parameters (e.g. estimated value, variation), we can use some known distributions to represent probability functions. Common choice could be normal distribution lots of phenomena have distribution similar to normal in practice. In our equations will be situations easy, because logarithm eliminate exponential function from Gaussian definition. For inner probability this works quite well in case of non-contrast CT data, because we have more or less homogeneous region. But for outer region or for contrast images is probability described by normal distribution insufficient. According to (Krajíček, 2007) we decided to represent probability in more precise way by normalized density histogram. This solution is more flexible, because we don t need to guess the type of a distribution, which best fit to our problem. In (Krajíček, 2007) were histograms computed in the beginning of segmentation and during algorithm were recalculated according to actual contour position in a slice. This is important for keeping histograms updated real distribution of density will not be so different. We chose different approach, we built histograms from training set. This mean that they don t reflect distribution in segmented image precisely, but instead show distribution in average case. Precision we lost by not incorporating histogram recalculation we try to compensate by different mechanism shape representation. Other probability functions are defined over 3D or 4D domain. We use 3D grid, where every element contains information needed to represent defined probability functions. Model Training We formally described principles of the statistical model used in our segmentation algorithm, but this model must be trained at first. We need precisely segmented training set to do that. Every sample in training set is formed from the binary mask and the original data. These data should be denoised or preprocessed by other filters, which will be applied on the input data before segmentation. This should assure that model will be trained for same type of data as we will process. Histogram Computing. Training of average normalized histograms is simple. For every sample we get two histograms the first is for region selected by mask and the second is for its supplement. So now we have two sets of histograms for inner regions and outer regions. We need normalized average histogram, so we simply do a summation of all histograms in each set. We obtain two super-histograms and we now normalize them. Normalized histogram sum to one over its domain we need this feature, because we want to use the result as probability density function. If we have small training set, we would consider normalized histogram smoothing to decrease influence of outliers. Train Data Alignment. We can t simply blend samples, because we will lost model precision. If we want combine samples with bigger precision, we need to align them at first. Complex registration is not needed, because we combine samples from different subjects and exact match is impossible anyway. We should use the same apparatus for aligning samples as we use for model alignment at the beginning of the segmentation process. We did few assumptions before we chose a set of transformations, which we use for alignment of kidney samples. The first thing is that kidneys from same body side are oriented similarly over the subjects. It means that we need to rotate samples only a little. For segmentation we obtain from user two points (kidney poles), so we can use the same points for setting transformation parameters. Training data can either contain coordinates of the organ poles or we can easily emulate user input by computing centers of gravity in two outer slices of training sample (kidneys have quite simple shape especially around poles). Another thing we should keep in mind is that we will use the model all the time during segmentation process. So we need to have all operations on it fast and simple. That is why we restricted transformation only to translation and skew. Kidneys are all oriented similarly in slices, so we omitted rotation around an axis perpendicular to slices. To alter the rest of the rotations we used skew transformation (parameters are easily computed from coordinates of the poles). Shape Model Training. When we have all training data aligned, we can proceed to shape model training. We will simply make average image of all aligned bit masks. If we assume that in every bit 73

5 mask meant 1 object and 0 background, value in averaged image give us probability of point being inside of the region. Supplement to 1 is probability of the point being outside of the region. We use these values and precompute logarithm of their ratio, which is used in region energy term. General Model Training. So we know how to train special parts of our statistical organ model, now we have to train general shape and density model. We use same procedure as for shape model, but computations in each element will be different. Every cell of model grid contains histograms for in probability, out probability and logarithm of their ratio. For shape model we averaged all aligned voxels from masks into one value, but for general shape/intensity model we add density values from aligned data voxels into cell histograms and we decide to which one by mask image. At the end, we normalize all cell histograms and compute their ratio logarithm. If we don t have enough training samples, we should consider histogram smoothing. Model Instances. For proper function of the model we need it to reflect real objects (kidneys in our case) very closely. To achieve that, we need to lower variation of the training data without losing representativity. First base is data alignment, which we already described. We can t proceed further without removing the most distinctive samples from training set but selective sample removal will certainly influence representativity. But there is a solution by reduction of the training set we defined new subclass of objects, so we use model trained from this reduced training set only for segmentation of the objects falling in this subclass. If we want to be able to successfully run segmentation on objects from the whole object class, and not only from its subset, we need to cover it by different subclasses and train separate model for each of them. For our case (kidney segmentation), we should have at least two models one for each side. But we will achieve better results by splitting problem to more categories. Not only for contrast and noncontrast but also for each distinguishable type of kidney shape (small, large, etc.). Trained user should easily decide the right category and choose appropriate model for segmentation. Number of categories depends on amount of the possible training data (danger of model undertraining for some small shape category), user friendliness (we don t want to flood user by large list of types) and precision of the models (if we have models without details, we don t need accurate categorization). For design of the object categories can be used some method from cluster analysis. Other Parts of the Model. We showed ways to model training, but during that process we can compute other statistics, which can be used for different purposes, other than application in energy terms. One possibility is computation of organ s point of gravity and weight in each layer. These values can be used in separate slice initialization version of sliced algorithm for preparation of curve s initial shape. In model can be possibly present whole average parametric initialization for snakes, which can be used instead of the initialization by circle. Similar statistics can be added to model for utilization in other segmentation methods, gradient image preprocessing, etc. 3D Sliced Segmentation Now we have finished 2D segmentation algorithm. For segmentation of 3D data we must execute algorithm in volume slices. We let user to specify two organ poles, which we use as region of interest definition and hint for alignment of model with actual dataset. Now we can proceed to execution of segmentation algorithm. In (Krajíček, 2007) algorithm started on those poles and used results obtained in each slice as initialization for the next one. We showed that we do not need precise initialization so we can start computations in each slice separately. Using initialization by neighboring slice, we can execute segmentation in two separate threads, each going from one pole ((Krajíček, 2007)). Separate initialization allows us completely independent computations in each slice. This means we can parallelize segmentation process on high level each slice can have its own thread. 74

6 a b c Figure 2. Separate slice initialization: 1 Poles set by user; 2 Computed initialization points for each slice; 3 Execute slice segmentation algorithm for separate initialization. Conclusion We successfully moved to statistical model most of the information needed by algorithm during its computation. This allowed us to use easy initialization of contour shape, reduce number of parameters controlling computations and most importantly we achieved to speed up original algorithm times (when executed in one thread). This makes segmentation of kidneys from thick slice CT series almost interactive when running in one thread (around 0.5s on common PC). On the other side we have to train precise representations of body organs, so we should have model for each kind of measurement (with or without contrast agent), demography group (kids, seniors, etc.) and pathology type to improve quality of results. References Adalsteinsson, D. and Sethian, J. A., A fast level set method for propagating interfaces, Journal of Computational Physics, 118, , Jacob, M., Blu, T., and Unser, M., Efficient Energies and Algorithms for Parametric Snakes, IEEE Transactions on Image Processing, 13, , URL Kass, M., Witkin, A., and Terzopoulos, D., Snakes: Active contour models, INTERNATIONAL JOUR- NAL OF COMPUTER VISION, 1, , Kolomaznk, J., Organ Segmentation, Master s thesis, Faculty of Mathematics and Physics, Charles University in Prague, Krajíček, V., Měření objemu v 3D datech, Master s thesis, Faculty of Mathematics and Physics, Charles University in Prague, McInerney, T. and Terzopoulos, D., Deformable models in medical image analysis: A survey, Medical Image Analysis, 1, , Orderud, F. and Rabben, S. I., Real-time 3d segmentation of the left ventricle using deformable subdivision surfaces, Computer Vision and Pattern Recognition, IEEE Computer Society Conference on, 0, 1 8, Straka, M., Cruz, A. L., Köchl, A., Srámek, M., Gröller, M. E., and Fleischmann, D., 3d watershed transform combined with a probabilistic atlas for medical image segmentation, in MIT 2003, pp. URL

Learning-based Neuroimage Registration

Learning-based Neuroimage Registration Learning-based Neuroimage Registration Leonid Teverovskiy and Yanxi Liu 1 October 2004 CMU-CALD-04-108, CMU-RI-TR-04-59 School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 Abstract

More information

Optimization. Intelligent Scissors (see also Snakes)

Optimization. Intelligent Scissors (see also Snakes) Optimization We can define a cost for possible solutions Number of solutions is large (eg., exponential) Efficient search is needed Global methods: cleverly find best solution without considering all.

More information

Image Segmentation II Advanced Approaches

Image Segmentation II Advanced Approaches Image Segmentation II Advanced Approaches Jorge Jara W. 1,2 1 Department of Computer Science DCC, U. of Chile 2 SCIAN-Lab, BNI Outline 1. Segmentation I Digital image processing Segmentation basics 2.

More information

Coarse-to-fine image registration

Coarse-to-fine image registration Today we will look at a few important topics in scale space in computer vision, in particular, coarseto-fine approaches, and the SIFT feature descriptor. I will present only the main ideas here to give

More information

MEDICAL IMAGE NOISE REDUCTION AND REGION CONTRAST ENHANCEMENT USING PARTIAL DIFFERENTIAL EQUATIONS

MEDICAL IMAGE NOISE REDUCTION AND REGION CONTRAST ENHANCEMENT USING PARTIAL DIFFERENTIAL EQUATIONS MEDICAL IMAGE NOISE REDUCTION AND REGION CONTRAST ENHANCEMENT USING PARTIAL DIFFERENTIAL EQUATIONS Miguel Alemán-Flores, Luis Álvarez-León Departamento de Informática y Sistemas, Universidad de Las Palmas

More information

Variational Methods II

Variational Methods II Mathematical Foundations of Computer Graphics and Vision Variational Methods II Luca Ballan Institute of Visual Computing Last Lecture If we have a topological vector space with an inner product and functionals

More information

Lecture 12 Level Sets & Parametric Transforms. sec & ch. 11 of Machine Vision by Wesley E. Snyder & Hairong Qi

Lecture 12 Level Sets & Parametric Transforms. sec & ch. 11 of Machine Vision by Wesley E. Snyder & Hairong Qi Lecture 12 Level Sets & Parametric Transforms sec. 8.5.2 & ch. 11 of Machine Vision by Wesley E. Snyder & Hairong Qi Spring 2017 16-725 (CMU RI) : BioE 2630 (Pitt) Dr. John Galeotti The content of these

More information

Chapter 9 Object Tracking an Overview

Chapter 9 Object Tracking an Overview Chapter 9 Object Tracking an Overview The output of the background subtraction algorithm, described in the previous chapter, is a classification (segmentation) of pixels into foreground pixels (those belonging

More information

Edge and local feature detection - 2. Importance of edge detection in computer vision

Edge and local feature detection - 2. Importance of edge detection in computer vision Edge and local feature detection Gradient based edge detection Edge detection by function fitting Second derivative edge detectors Edge linking and the construction of the chain graph Edge and local feature

More information

3D Surface Reconstruction of the Brain based on Level Set Method

3D Surface Reconstruction of the Brain based on Level Set Method 3D Surface Reconstruction of the Brain based on Level Set Method Shijun Tang, Bill P. Buckles, and Kamesh Namuduri Department of Computer Science & Engineering Department of Electrical Engineering University

More information

Segmentation. Namrata Vaswani,

Segmentation. Namrata Vaswani, Segmentation Namrata Vaswani, namrata@iastate.edu Read Sections 5.1,5.2,5.3 of [1] Edge detection and filtering : Canny edge detection algorithm to get a contour of the object boundary Hough transform:

More information

For Information on SNAKEs. Active Contours (SNAKES) Improve Boundary Detection. Back to boundary detection. This is non-parametric

For Information on SNAKEs. Active Contours (SNAKES) Improve Boundary Detection. Back to boundary detection. This is non-parametric Active Contours (SNAKES) Back to boundary detection This time using perceptual grouping. This is non-parametric We re not looking for a contour of a specific shape. Just a good contour. For Information

More information

CS 231A Computer Vision (Fall 2012) Problem Set 3

CS 231A Computer Vision (Fall 2012) Problem Set 3 CS 231A Computer Vision (Fall 2012) Problem Set 3 Due: Nov. 13 th, 2012 (2:15pm) 1 Probabilistic Recursion for Tracking (20 points) In this problem you will derive a method for tracking a point of interest

More information

Image Segmentation. Ross Whitaker SCI Institute, School of Computing University of Utah

Image Segmentation. Ross Whitaker SCI Institute, School of Computing University of Utah Image Segmentation Ross Whitaker SCI Institute, School of Computing University of Utah What is Segmentation? Partitioning images/volumes into meaningful pieces Partitioning problem Labels Isolating a specific

More information

Image Processing. Filtering. Slide 1

Image Processing. Filtering. Slide 1 Image Processing Filtering Slide 1 Preliminary Image generation Original Noise Image restoration Result Slide 2 Preliminary Classic application: denoising However: Denoising is much more than a simple

More information

Matching. Compare region of image to region of image. Today, simplest kind of matching. Intensities similar.

Matching. Compare region of image to region of image. Today, simplest kind of matching. Intensities similar. Matching Compare region of image to region of image. We talked about this for stereo. Important for motion. Epipolar constraint unknown. But motion small. Recognition Find object in image. Recognize object.

More information

Image Segmentation and Registration

Image Segmentation and Registration Image Segmentation and Registration Dr. Christine Tanner (tanner@vision.ee.ethz.ch) Computer Vision Laboratory, ETH Zürich Dr. Verena Kaynig, Machine Learning Laboratory, ETH Zürich Outline Segmentation

More information

Panoramic Image Stitching

Panoramic Image Stitching Mcgill University Panoramic Image Stitching by Kai Wang Pengbo Li A report submitted in fulfillment for the COMP 558 Final project in the Faculty of Computer Science April 2013 Mcgill University Abstract

More information

Level Set Evolution without Reinitilization

Level Set Evolution without Reinitilization Level Set Evolution without Reinitilization Outline Parametric active contour (snake) models. Concepts of Level set method and geometric active contours. A level set formulation without reinitialization.

More information

Blood Vessel Visualization on CT Data

Blood Vessel Visualization on CT Data WDS'12 Proceedings of Contributed Papers, Part I, 88 93, 2012. ISBN 978-80-7378-224-5 MATFYZPRESS Blood Vessel Visualization on CT Data J. Dupej Charles University Prague, Faculty of Mathematics and Physics,

More information

How to print a Hypercube

How to print a Hypercube How to print a Hypercube Henry Segerman One of the things that mathematics is about, perhaps the thing that mathematics is about, is trying to make things easier to understand. John von Neumann once said

More information

Dr. Ulas Bagci

Dr. Ulas Bagci Lecture 9: Deformable Models and Segmentation CAP-Computer Vision Lecture 9-Deformable Models and Segmentation Dr. Ulas Bagci bagci@ucf.edu Lecture 9: Deformable Models and Segmentation Motivation A limitation

More information

1. Two double lectures about deformable contours. 4. The transparencies define the exam requirements. 1. Matlab demonstration

1. Two double lectures about deformable contours. 4. The transparencies define the exam requirements. 1. Matlab demonstration Practical information INF 5300 Deformable contours, I An introduction 1. Two double lectures about deformable contours. 2. The lectures are based on articles, references will be given during the course.

More information

Automatic Grayscale Classification using Histogram Clustering for Active Contour Models

Automatic Grayscale Classification using Histogram Clustering for Active Contour Models Research Article International Journal of Current Engineering and Technology ISSN 2277-4106 2013 INPRESSCO. All Rights Reserved. Available at http://inpressco.com/category/ijcet Automatic Grayscale Classification

More information

CS 223B Computer Vision Problem Set 3

CS 223B Computer Vision Problem Set 3 CS 223B Computer Vision Problem Set 3 Due: Feb. 22 nd, 2011 1 Probabilistic Recursion for Tracking In this problem you will derive a method for tracking a point of interest through a sequence of images.

More information

3 Nonlinear Regression

3 Nonlinear Regression CSC 4 / CSC D / CSC C 3 Sometimes linear models are not sufficient to capture the real-world phenomena, and thus nonlinear models are necessary. In regression, all such models will have the same basic

More information

Digital Image Processing. Prof. P.K. Biswas. Department of Electronics & Electrical Communication Engineering

Digital Image Processing. Prof. P.K. Biswas. Department of Electronics & Electrical Communication Engineering Digital Image Processing Prof. P.K. Biswas Department of Electronics & Electrical Communication Engineering Indian Institute of Technology, Kharagpur Image Segmentation - III Lecture - 31 Hello, welcome

More information

Problem Solving Assignment 1

Problem Solving Assignment 1 CS6240 Problem Solving Assignment 1 p. 1/20 Problem Solving Assignment 1 CS6240 Multimedia Analysis Daniel Dahlmeier National University of Singapore CS6240 Problem Solving Assignment 1 p. 2/20 Introduction

More information

Segmentation. Separate image into coherent regions

Segmentation. Separate image into coherent regions Segmentation II Segmentation Separate image into coherent regions Berkeley segmentation database: http://www.eecs.berkeley.edu/research/projects/cs/vision/grouping/segbench/ Slide by L. Lazebnik Interactive

More information

ADAPTIVE GRAPH CUTS WITH TISSUE PRIORS FOR BRAIN MRI SEGMENTATION

ADAPTIVE GRAPH CUTS WITH TISSUE PRIORS FOR BRAIN MRI SEGMENTATION ADAPTIVE GRAPH CUTS WITH TISSUE PRIORS FOR BRAIN MRI SEGMENTATION Abstract: MIP Project Report Spring 2013 Gaurav Mittal 201232644 This is a detailed report about the course project, which was to implement

More information

Robust Lip Contour Extraction using Separability of Multi-Dimensional Distributions

Robust Lip Contour Extraction using Separability of Multi-Dimensional Distributions Robust Lip Contour Extraction using Separability of Multi-Dimensional Distributions Tomokazu Wakasugi, Masahide Nishiura and Kazuhiro Fukui Corporate Research and Development Center, Toshiba Corporation

More information

CHAPTER 2. Morphometry on rodent brains. A.E.H. Scheenstra J. Dijkstra L. van der Weerd

CHAPTER 2. Morphometry on rodent brains. A.E.H. Scheenstra J. Dijkstra L. van der Weerd CHAPTER 2 Morphometry on rodent brains A.E.H. Scheenstra J. Dijkstra L. van der Weerd This chapter was adapted from: Volumetry and other quantitative measurements to assess the rodent brain, In vivo NMR

More information

Clustering Part 4 DBSCAN

Clustering Part 4 DBSCAN Clustering Part 4 Dr. Sanjay Ranka Professor Computer and Information Science and Engineering University of Florida, Gainesville DBSCAN DBSCAN is a density based clustering algorithm Density = number of

More information

Unsupervised learning in Vision

Unsupervised learning in Vision Chapter 7 Unsupervised learning in Vision The fields of Computer Vision and Machine Learning complement each other in a very natural way: the aim of the former is to extract useful information from visual

More information

Grid-Based Genetic Algorithm Approach to Colour Image Segmentation

Grid-Based Genetic Algorithm Approach to Colour Image Segmentation Grid-Based Genetic Algorithm Approach to Colour Image Segmentation Marco Gallotta Keri Woods Supervised by Audrey Mbogho Image Segmentation Identifying and extracting distinct, homogeneous regions from

More information

CAP 5415 Computer Vision Fall 2012

CAP 5415 Computer Vision Fall 2012 CAP 5415 Computer Vision Fall 01 Dr. Mubarak Shah Univ. of Central Florida Office 47-F HEC Lecture-5 SIFT: David Lowe, UBC SIFT - Key Point Extraction Stands for scale invariant feature transform Patented

More information

3D Models and Matching

3D Models and Matching 3D Models and Matching representations for 3D object models particular matching techniques alignment-based systems appearance-based systems GC model of a screwdriver 1 3D Models Many different representations

More information

HOUGH TRANSFORM CS 6350 C V

HOUGH TRANSFORM CS 6350 C V HOUGH TRANSFORM CS 6350 C V HOUGH TRANSFORM The problem: Given a set of points in 2-D, find if a sub-set of these points, fall on a LINE. Hough Transform One powerful global method for detecting edges

More information

Modern Medical Image Analysis 8DC00 Exam

Modern Medical Image Analysis 8DC00 Exam Parts of answers are inside square brackets [... ]. These parts are optional. Answers can be written in Dutch or in English, as you prefer. You can use drawings and diagrams to support your textual answers.

More information

IMAGE SEGMENTATION. Václav Hlaváč

IMAGE SEGMENTATION. Václav Hlaváč IMAGE SEGMENTATION Václav Hlaváč Czech Technical University in Prague Faculty of Electrical Engineering, Department of Cybernetics Center for Machine Perception http://cmp.felk.cvut.cz/ hlavac, hlavac@fel.cvut.cz

More information

SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS

SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS Cognitive Robotics Original: David G. Lowe, 004 Summary: Coen van Leeuwen, s1460919 Abstract: This article presents a method to extract

More information

3D Models and Matching

3D Models and Matching 3D Models and Matching representations for 3D object models particular matching techniques alignment-based systems appearance-based systems GC model of a screwdriver 1 3D Models Many different representations

More information

Digital Image Processing

Digital Image Processing Digital Image Processing Part 9: Representation and Description AASS Learning Systems Lab, Dep. Teknik Room T1209 (Fr, 11-12 o'clock) achim.lilienthal@oru.se Course Book Chapter 11 2011-05-17 Contents

More information

University of Florida CISE department Gator Engineering. Clustering Part 4

University of Florida CISE department Gator Engineering. Clustering Part 4 Clustering Part 4 Dr. Sanjay Ranka Professor Computer and Information Science and Engineering University of Florida, Gainesville DBSCAN DBSCAN is a density based clustering algorithm Density = number of

More information

Local Features: Detection, Description & Matching

Local Features: Detection, Description & Matching Local Features: Detection, Description & Matching Lecture 08 Computer Vision Material Citations Dr George Stockman Professor Emeritus, Michigan State University Dr David Lowe Professor, University of British

More information

Lecture 15: Segmentation (Edge Based, Hough Transform)

Lecture 15: Segmentation (Edge Based, Hough Transform) Lecture 15: Segmentation (Edge Based, Hough Transform) c Bryan S. Morse, Brigham Young University, 1998 000 Last modified on February 3, 000 at :00 PM Contents 15.1 Introduction..............................................

More information

Points Lines Connected points X-Y Scatter. X-Y Matrix Star Plot Histogram Box Plot. Bar Group Bar Stacked H-Bar Grouped H-Bar Stacked

Points Lines Connected points X-Y Scatter. X-Y Matrix Star Plot Histogram Box Plot. Bar Group Bar Stacked H-Bar Grouped H-Bar Stacked Plotting Menu: QCExpert Plotting Module graphs offers various tools for visualization of uni- and multivariate data. Settings and options in different types of graphs allow for modifications and customizations

More information

Level-set MCMC Curve Sampling and Geometric Conditional Simulation

Level-set MCMC Curve Sampling and Geometric Conditional Simulation Level-set MCMC Curve Sampling and Geometric Conditional Simulation Ayres Fan John W. Fisher III Alan S. Willsky February 16, 2007 Outline 1. Overview 2. Curve evolution 3. Markov chain Monte Carlo 4. Curve

More information

Adaptive active contours (snakes) for the segmentation of complex structures in biological images

Adaptive active contours (snakes) for the segmentation of complex structures in biological images Adaptive active contours (snakes) for the segmentation of complex structures in biological images Philippe Andrey a and Thomas Boudier b a Analyse et Modélisation en Imagerie Biologique, Laboratoire Neurobiologie

More information

EE795: Computer Vision and Intelligent Systems

EE795: Computer Vision and Intelligent Systems EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 WRI C225 Lecture 04 130131 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Histogram Equalization Image Filtering Linear

More information

Object Recognition Using Pictorial Structures. Daniel Huttenlocher Computer Science Department. In This Talk. Object recognition in computer vision

Object Recognition Using Pictorial Structures. Daniel Huttenlocher Computer Science Department. In This Talk. Object recognition in computer vision Object Recognition Using Pictorial Structures Daniel Huttenlocher Computer Science Department Joint work with Pedro Felzenszwalb, MIT AI Lab In This Talk Object recognition in computer vision Brief definition

More information

Scientific Visualization Example exam questions with commented answers

Scientific Visualization Example exam questions with commented answers Scientific Visualization Example exam questions with commented answers The theoretical part of this course is evaluated by means of a multiple- choice exam. The questions cover the material mentioned during

More information

Image Processing

Image Processing Image Processing 159.731 Canny Edge Detection Report Syed Irfanullah, Azeezullah 00297844 Danh Anh Huynh 02136047 1 Canny Edge Detection INTRODUCTION Edges Edges characterize boundaries and are therefore

More information

Unstructured Mesh Generation for Implicit Moving Geometries and Level Set Applications

Unstructured Mesh Generation for Implicit Moving Geometries and Level Set Applications Unstructured Mesh Generation for Implicit Moving Geometries and Level Set Applications Per-Olof Persson (persson@mit.edu) Department of Mathematics Massachusetts Institute of Technology http://www.mit.edu/

More information

Feature Detectors and Descriptors: Corners, Lines, etc.

Feature Detectors and Descriptors: Corners, Lines, etc. Feature Detectors and Descriptors: Corners, Lines, etc. Edges vs. Corners Edges = maxima in intensity gradient Edges vs. Corners Corners = lots of variation in direction of gradient in a small neighborhood

More information

6. Object Identification L AK S H M O U. E D U

6. Object Identification L AK S H M O U. E D U 6. Object Identification L AK S H M AN @ O U. E D U Objects Information extracted from spatial grids often need to be associated with objects not just an individual pixel Group of pixels that form a real-world

More information

Edge detection. Convert a 2D image into a set of curves. Extracts salient features of the scene More compact than pixels

Edge detection. Convert a 2D image into a set of curves. Extracts salient features of the scene More compact than pixels Edge Detection Edge detection Convert a 2D image into a set of curves Extracts salient features of the scene More compact than pixels Origin of Edges surface normal discontinuity depth discontinuity surface

More information

K-Means Clustering Using Localized Histogram Analysis

K-Means Clustering Using Localized Histogram Analysis K-Means Clustering Using Localized Histogram Analysis Michael Bryson University of South Carolina, Department of Computer Science Columbia, SC brysonm@cse.sc.edu Abstract. The first step required for many

More information

Image Segmentation. Ross Whitaker SCI Institute, School of Computing University of Utah

Image Segmentation. Ross Whitaker SCI Institute, School of Computing University of Utah Image Segmentation Ross Whitaker SCI Institute, School of Computing University of Utah What is Segmentation? Partitioning images/volumes into meaningful pieces Partitioning problem Labels Isolating a specific

More information

Snakes reparameterization for noisy images segmentation and targets tracking

Snakes reparameterization for noisy images segmentation and targets tracking Snakes reparameterization for noisy images segmentation and targets tracking Idrissi Sidi Yassine, Samir Belfkih. Lycée Tawfik Elhakim Zawiya de Noaceur, route de Marrakech, Casablanca, maroc. Laboratoire

More information

How does bilateral filter relates with other methods?

How does bilateral filter relates with other methods? A Gentle Introduction to Bilateral Filtering and its Applications How does bilateral filter relates with other methods? Fredo Durand (MIT CSAIL) Slides by Pierre Kornprobst (INRIA) 0:35 Many people worked

More information

Multimedia Computing: Algorithms, Systems, and Applications: Edge Detection

Multimedia Computing: Algorithms, Systems, and Applications: Edge Detection Multimedia Computing: Algorithms, Systems, and Applications: Edge Detection By Dr. Yu Cao Department of Computer Science The University of Massachusetts Lowell Lowell, MA 01854, USA Part of the slides

More information

EEM 463 Introduction to Image Processing. Week 3: Intensity Transformations

EEM 463 Introduction to Image Processing. Week 3: Intensity Transformations EEM 463 Introduction to Image Processing Week 3: Intensity Transformations Fall 2013 Instructor: Hatice Çınar Akakın, Ph.D. haticecinarakakin@anadolu.edu.tr Anadolu University Enhancement Domains Spatial

More information

Medical Image Segmentation Based on Mutual Information Maximization

Medical Image Segmentation Based on Mutual Information Maximization Medical Image Segmentation Based on Mutual Information Maximization J.Rigau, M.Feixas, M.Sbert, A.Bardera, and I.Boada Institut d Informatica i Aplicacions, Universitat de Girona, Spain {jaume.rigau,miquel.feixas,mateu.sbert,anton.bardera,imma.boada}@udg.es

More information

Other Linear Filters CS 211A

Other Linear Filters CS 211A Other Linear Filters CS 211A Slides from Cornelia Fermüller and Marc Pollefeys Edge detection Convert a 2D image into a set of curves Extracts salient features of the scene More compact than pixels Origin

More information

2D image segmentation based on spatial coherence

2D image segmentation based on spatial coherence 2D image segmentation based on spatial coherence Václav Hlaváč Czech Technical University in Prague Center for Machine Perception (bridging groups of the) Czech Institute of Informatics, Robotics and Cybernetics

More information

The goal is the definition of points with numbers and primitives with equations or functions. The definition of points with numbers requires a

The goal is the definition of points with numbers and primitives with equations or functions. The definition of points with numbers requires a The goal is the definition of points with numbers and primitives with equations or functions. The definition of points with numbers requires a coordinate system and then the measuring of the point with

More information

Announcements. Edges. Last Lecture. Gradients: Numerical Derivatives f(x) Edge Detection, Lines. Intro Computer Vision. CSE 152 Lecture 10

Announcements. Edges. Last Lecture. Gradients: Numerical Derivatives f(x) Edge Detection, Lines. Intro Computer Vision. CSE 152 Lecture 10 Announcements Assignment 2 due Tuesday, May 4. Edge Detection, Lines Midterm: Thursday, May 6. Introduction to Computer Vision CSE 152 Lecture 10 Edges Last Lecture 1. Object boundaries 2. Surface normal

More information

The VesselGlyph: Focus & Context Visualization in CT-Angiography

The VesselGlyph: Focus & Context Visualization in CT-Angiography The VesselGlyph: Focus & Context Visualization in CT-Angiography Matúš Straka M. Šrámek, A. La Cruz E. Gröller, D. Fleischmann Contents Motivation:» Why again a new visualization method for vessel data?

More information

Images from 3D Creative Magazine. 3D Modelling Systems

Images from 3D Creative Magazine. 3D Modelling Systems Images from 3D Creative Magazine 3D Modelling Systems Contents Reference & Accuracy 3D Primitives Transforms Move (Translate) Rotate Scale Mirror Align 3D Booleans Deforms Bend Taper Skew Twist Squash

More information

#$ % $ $& "$%% " $ '$ " '

#$ % $ $& $%%  $ '$  ' ! " This section of the course covers techniques for pairwise (i.e., scanto-scan) and global (i.e., involving more than 2 scans) alignment, given that the algorithms are constrained to obtain a rigid-body

More information

Lecture 16: Computer Vision

Lecture 16: Computer Vision CS4442/9542b: Artificial Intelligence II Prof. Olga Veksler Lecture 16: Computer Vision Motion Slides are from Steve Seitz (UW), David Jacobs (UMD) Outline Motion Estimation Motion Field Optical Flow Field

More information

Lecture 16: Computer Vision

Lecture 16: Computer Vision CS442/542b: Artificial ntelligence Prof. Olga Veksler Lecture 16: Computer Vision Motion Slides are from Steve Seitz (UW), David Jacobs (UMD) Outline Motion Estimation Motion Field Optical Flow Field Methods

More information

Bayesian Methods in Vision: MAP Estimation, MRFs, Optimization

Bayesian Methods in Vision: MAP Estimation, MRFs, Optimization Bayesian Methods in Vision: MAP Estimation, MRFs, Optimization CS 650: Computer Vision Bryan S. Morse Optimization Approaches to Vision / Image Processing Recurring theme: Cast vision problem as an optimization

More information

EE368 Project Report CD Cover Recognition Using Modified SIFT Algorithm

EE368 Project Report CD Cover Recognition Using Modified SIFT Algorithm EE368 Project Report CD Cover Recognition Using Modified SIFT Algorithm Group 1: Mina A. Makar Stanford University mamakar@stanford.edu Abstract In this report, we investigate the application of the Scale-Invariant

More information

COMPUTER AND ROBOT VISION

COMPUTER AND ROBOT VISION VOLUME COMPUTER AND ROBOT VISION Robert M. Haralick University of Washington Linda G. Shapiro University of Washington A^ ADDISON-WESLEY PUBLISHING COMPANY Reading, Massachusetts Menlo Park, California

More information

Contrast Optimization A new way to optimize performance Kenneth Moore, Technical Fellow

Contrast Optimization A new way to optimize performance Kenneth Moore, Technical Fellow Contrast Optimization A new way to optimize performance Kenneth Moore, Technical Fellow What is Contrast Optimization? Contrast Optimization (CO) is a new technique for improving performance of imaging

More information

Morphometric Analysis of Biomedical Images. Sara Rolfe 10/9/17

Morphometric Analysis of Biomedical Images. Sara Rolfe 10/9/17 Morphometric Analysis of Biomedical Images Sara Rolfe 10/9/17 Morphometric Analysis of Biomedical Images Object surface contours Image difference features Compact representation of feature differences

More information

Edges and Lines Readings: Chapter 10: better edge detectors line finding circle finding

Edges and Lines Readings: Chapter 10: better edge detectors line finding circle finding Edges and Lines Readings: Chapter 10: 10.2.3-10.3 better edge detectors line finding circle finding 1 Lines and Arcs Segmentation In some image sets, lines, curves, and circular arcs are more useful than

More information

Snakes, level sets and graphcuts. (Deformable models)

Snakes, level sets and graphcuts. (Deformable models) INSTITUTE OF INFORMATION AND COMMUNICATION TECHNOLOGIES BULGARIAN ACADEMY OF SCIENCE Snakes, level sets and graphcuts (Deformable models) Centro de Visión por Computador, Departament de Matemàtica Aplicada

More information

Today. Lecture 4: Last time. The EM algorithm. We examine clustering in a little more detail; we went over it a somewhat quickly last time

Today. Lecture 4: Last time. The EM algorithm. We examine clustering in a little more detail; we went over it a somewhat quickly last time Today Lecture 4: We examine clustering in a little more detail; we went over it a somewhat quickly last time The CAD data will return and give us an opportunity to work with curves (!) We then examine

More information

Iso-surface cell search. Iso-surface Cells. Efficient Searching. Efficient search methods. Efficient iso-surface cell search. Problem statement:

Iso-surface cell search. Iso-surface Cells. Efficient Searching. Efficient search methods. Efficient iso-surface cell search. Problem statement: Iso-Contouring Advanced Issues Iso-surface cell search 1. Efficiently determining which cells to examine. 2. Using iso-contouring as a slicing mechanism 3. Iso-contouring in higher dimensions 4. Texturing

More information

Tips and Guidance for Analyzing Data. Executive Summary

Tips and Guidance for Analyzing Data. Executive Summary Tips and Guidance for Analyzing Data Executive Summary This document has information and suggestions about three things: 1) how to quickly do a preliminary analysis of time-series data; 2) key things to

More information

3 Nonlinear Regression

3 Nonlinear Regression 3 Linear models are often insufficient to capture the real-world phenomena. That is, the relation between the inputs and the outputs we want to be able to predict are not linear. As a consequence, nonlinear

More information

Implementation and Comparison of Feature Detection Methods in Image Mosaicing

Implementation and Comparison of Feature Detection Methods in Image Mosaicing IOSR Journal of Electronics and Communication Engineering (IOSR-JECE) e-issn: 2278-2834,p-ISSN: 2278-8735 PP 07-11 www.iosrjournals.org Implementation and Comparison of Feature Detection Methods in Image

More information

Image Registration Lecture 4: First Examples

Image Registration Lecture 4: First Examples Image Registration Lecture 4: First Examples Prof. Charlene Tsai Outline Example Intensity-based registration SSD error function Image mapping Function minimization: Gradient descent Derivative calculation

More information

Topology and Topological Spaces

Topology and Topological Spaces Topology and Topological Spaces Mathematical spaces such as vector spaces, normed vector spaces (Banach spaces), and metric spaces are generalizations of ideas that are familiar in R or in R n. For example,

More information

weighted minimal surface model for surface reconstruction from scattered points, curves, and/or pieces of surfaces.

weighted minimal surface model for surface reconstruction from scattered points, curves, and/or pieces of surfaces. weighted minimal surface model for surface reconstruction from scattered points, curves, and/or pieces of surfaces. joint work with (S. Osher, R. Fedkiw and M. Kang) Desired properties for surface reconstruction:

More information

2 Michael E. Leventon and Sarah F. F. Gibson a b c d Fig. 1. (a, b) Two MR scans of a person's knee. Both images have high resolution in-plane, but ha

2 Michael E. Leventon and Sarah F. F. Gibson a b c d Fig. 1. (a, b) Two MR scans of a person's knee. Both images have high resolution in-plane, but ha Model Generation from Multiple Volumes using Constrained Elastic SurfaceNets Michael E. Leventon and Sarah F. F. Gibson 1 MIT Artificial Intelligence Laboratory, Cambridge, MA 02139, USA leventon@ai.mit.edu

More information

Various Methods for Medical Image Segmentation

Various Methods for Medical Image Segmentation Various Methods for Medical Image Segmentation From Level Set to Convex Relaxation Doyeob Yeo and Soomin Jeon Computational Mathematics and Imaging Lab. Department of Mathematical Sciences, KAIST Hansang

More information

Registration Techniques

Registration Techniques EMBO Practical Course on Light Sheet Microscopy Junior-Prof. Dr. Olaf Ronneberger Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany O. Ronneberger,

More information

Automatic Logo Detection and Removal

Automatic Logo Detection and Removal Automatic Logo Detection and Removal Miriam Cha, Pooya Khorrami and Matthew Wagner Electrical and Computer Engineering Carnegie Mellon University Pittsburgh, PA 15213 {mcha,pkhorrami,mwagner}@ece.cmu.edu

More information

Implicit Active Shape Models for 3D Segmentation in MR Imaging

Implicit Active Shape Models for 3D Segmentation in MR Imaging Implicit Active Shape Models for 3D Segmentation in MR Imaging Mikaël Rousson 1, Nikos Paragios 2, and Rachid Deriche 1 1 I.N.R.I.A. Sophia Antipolis, France E-mail: {Mikael.Rousson,Rachid.Deriche}@sophia.inria.fr

More information

How Do We Measure Protein Shape? A Pattern Matching Example. A Simple Pattern Matching Algorithm. Comparing Protein Structures II

How Do We Measure Protein Shape? A Pattern Matching Example. A Simple Pattern Matching Algorithm. Comparing Protein Structures II How Do We Measure Protein Shape? omparing Protein Structures II Protein function is largely based on the proteins geometric shape Protein substructures with similar shapes are likely to share a common

More information

6. Applications - Text recognition in videos - Semantic video analysis

6. Applications - Text recognition in videos - Semantic video analysis 6. Applications - Text recognition in videos - Semantic video analysis Stephan Kopf 1 Motivation Goal: Segmentation and classification of characters Only few significant features are visible in these simple

More information

Computer vision: models, learning and inference. Chapter 13 Image preprocessing and feature extraction

Computer vision: models, learning and inference. Chapter 13 Image preprocessing and feature extraction Computer vision: models, learning and inference Chapter 13 Image preprocessing and feature extraction Preprocessing The goal of pre-processing is to try to reduce unwanted variation in image due to lighting,

More information

Leow Wee Kheng CS4243 Computer Vision and Pattern Recognition. Motion Tracking. CS4243 Motion Tracking 1

Leow Wee Kheng CS4243 Computer Vision and Pattern Recognition. Motion Tracking. CS4243 Motion Tracking 1 Leow Wee Kheng CS4243 Computer Vision and Pattern Recognition Motion Tracking CS4243 Motion Tracking 1 Changes are everywhere! CS4243 Motion Tracking 2 Illumination change CS4243 Motion Tracking 3 Shape

More information

Template Matching Rigid Motion

Template Matching Rigid Motion Template Matching Rigid Motion Find transformation to align two images. Focus on geometric features (not so much interesting with intensity images) Emphasis on tricks to make this efficient. Problem Definition

More information

Outline 7/2/201011/6/

Outline 7/2/201011/6/ Outline Pattern recognition in computer vision Background on the development of SIFT SIFT algorithm and some of its variations Computational considerations (SURF) Potential improvement Summary 01 2 Pattern

More information

An Automated Image-based Method for Multi-Leaf Collimator Positioning Verification in Intensity Modulated Radiation Therapy

An Automated Image-based Method for Multi-Leaf Collimator Positioning Verification in Intensity Modulated Radiation Therapy An Automated Image-based Method for Multi-Leaf Collimator Positioning Verification in Intensity Modulated Radiation Therapy Chenyang Xu 1, Siemens Corporate Research, Inc., Princeton, NJ, USA Xiaolei Huang,

More information