New version page

Evaluation of Binning Strategies

This preview shows page 1-2-3-4 out of 12 pages.

View Full Document
View Full Document

End of preview. Want to read all 12 pages?

Upload your study docs or become a GradeBuddy member to access this document.

View Full Document
Unformatted text preview:

Evaluation of Binning Strategies for Tissue Classification in Computed Tomography Images Stefanie Handricka, Bahare Naimipourb, Daniela Raicuc, Jacob Furstc aArizona State University, Tempe, AZ, USA, 85287 bUniversity of Illinois, Chicago, IL, USA, 60607 cIntelligent Multimedia Processing Laboratory School of Computer Science, Telecommunications, and Information Systems DePaul University, Chicago, IL, USA, 60604 ABSTRACT♠♠♠♠ Binning strategies have been used in much research work for image compression, feature extraction, classification, segmentation and other tasks, but rarely is there any rigorous investigation into which binning strategy is the best. Binning becomes a "hidden parameter" of the research method. This work rigorously investigates the results of three different binning strategies, linear binning, clipped binning, and nonlinear binning, for co-occurrence texture-based classification of the backbone, liver, heart, renal, and splenic parenchyma in high-resolution DICOM Computed Tomography (CT) images of the human chest and abdomen. Linear binning divides the gray-level range of [0..4095] into k1 equally sized bins, while clipped binning allocates one large bin for low intensity gray-levels [0..855] (air), one for higher intensities [1368..4095] (bone), and k2 equally sized bins for the soft tissues between [856..1368]. Nonlinear binning divides the gray-level range of [0..4095] into k3 bins of different sizes. These bins are further used to calculate the co-occurrence statistical model and its ten Haralick descriptors for texture quantification of gray-level images. The results of the texture quantification using each one of the three strategies and for different values of k1, k2 and k3 are evaluated with respect to their discrimination power using a decision tree classification algorithm and four classification performance metrics (sensitivity, specificity, precision and accuracy). Our preliminary results obtained on 1368 segmented DICOM images show that the optimal number of gray-levels is equal to 128 for linear binning, 512 for clipped binning, , and 256 for non-linear binning. Furthermore, when comparing the results of the three approaches, the nonlinear binning approach shows significant improvement for heart and spleen. Keywords: Binning, texture, co-occurrence matrix, computed tomography 1. INTRODUCTION Binning is a term used to describe the reduction of gray level intensities in an image, in this case by combining several intensity levels into a single intensity level, or bin. Linear binning describes the process of equally dividing the original gray level range of the image into k bins. Clipped binning describes the process of determining a range of interest, where the gray levels are equally divided into k bins, while the intensities above the desired range are placed into the highest bin, and the intensities below the desired range are placed into the lowest bin. Nonlinear binning describes the process of using histogram data to assign the original gray level range into possibly unequal bins. In this paper we evaluate three types of gray-level binning strategies, linear binning, clipped binning, and nonlinear binning, applied for efficient and accurate texture characterization of normal tissues in Computed Tomography (CT) of the chest and abdomen. The gray-level binning strategies serve as an image preprocessing step for the computation of co-occurrence matrices and Haralick texture features. The objective of this paper is to identify the binning method that reduces the gray-level resolution and the data redundancy present in the image while optimizing the classification ♠ This material is based upon work supported by the National Science Foundation under Grant No. 0453456.accuracy of five healthy tissues (backbone, liver, heart, renal, and splenic parenchyma ) based on ten Haralick texture descriptors. 2. BACKGROUND Previous work on the topic of binning has been done to determine an optimal k value using entropy and information gain8. Entropy is a measure of the information content of an image, and can be used to determine how many gray levels are necessary to represent all of the information11. Other work in the medical field evaluates a type of histogram binning compared to linear binning with respect to the speed and accuracy of pixel co-registration in magnetic resonance imaging (MRI)1. A method of classifying different lung disease as well as normal lung tissue utilized 16 gray levels but gave no explanation for choosing this particular k value2. 3. METHODOLOGY 3.1 The data set Our data set consisted of 1368 segmented DICOM images. There were 141 chest and abdomen high resolution CT images from two healthy patients. The five organs (backbone, liver, heart, renal, and splenic parenchyma) were then segmented from each of these CT images, resulting in around 340 segmented images for the group of organs. In order to extract more information from each organ and ultimately build a better classifier, each of the images was divided into quadrants resulting in the 1376 images. Eight images were removed because too much background data was present in these images and the texture data were not useful. The tissues represented in these images are all normal tissues, because if normal tissues can be classified correctly, then abnormal tissues would be easily identified. 3.2 Definition of clipped, linear, and nonlinear binning Linear binning divides the gray-level range of [0..4095] into k1 equally sized bins, clipped binning allocates one large bin for low intensity gray-levels [0..855] (air), one for higher intensities [1368..4095] (bone), and k2 equally sized bins for the soft tissues between [856..1368], and nonlinear binning distributes the gray levels into unequal bins based on histogram data. All types of binning are evaluated for five different k values: 32, 64, 128, 256, and 512; for nonlinear binning, this k value represents the number of clusters to be used in the k-means clustering9 approach, and does not represent the final number of gray levels in the image8. 3.3 Definition of the Co-occurrence Matrix The co-occurrence matrix is often used in texture analysis. By specifying a displacement vector, d, and counting all pairs of pixels separated by d with gray level intensities i and j, this co-occurrence matrix is created6. An image with 512 gray levels will have a 512 x 512 co-occurrence matrix,


Loading Unlocking...
Login

Join to view Evaluation of Binning Strategies and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Evaluation of Binning Strategies and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?