Tumor detection by using Zernike moments on segmented magnetic resonance brain images

Abstract

In this study, a novel method is proposed for the detection of tumor in magnetic resonance (MR) brain images. The performance of the novel method is investigated on one phantom and 20 original MR brain images with tumor and 50 normal (healthy) MR brain images.

Before the segmentation process, 2D continuous wavelet transform (CWT) is applied to reveal the characteristics of tissues in MR head images. Then, each MR image is segmented into seven classes (six head tissues and the background) by using the incremental supervised neural network (ISNN) and the wavelet-bands. After the segmentation process, the head is extracted from the background by simply discarding the background pixels. Symmetry axis of the head in the MR image is determined by using moment properties. Asymmetry is analyzed by using the Zernike moments of each of six tissues segmented in the head: two vectors are individually formed for the left and right hand sides of the symmetry axis on the sagittal plane by using the Zernike moments of the segmented tissues in the head. Presence of asymmetry and the tumors are inquired by considering the distance between these two vectors.

The performance of the proposed method is further investigated by moving the location of the tumor and by modifying its size in the phantom image. It is observed that tumor detection is successfully realized for the tumorous 20 MR brain images.

Introduction

This paper presents a general framework for analyzing structural and radiometric asymmetry in brain images and determining the existence of tumors. In a healthy brain, the left and right hemispheres are largely symmetric across the mid-sagittal plane. Brain tumors may belong to one or both of the following categories: mass-effect, in which the diseased tissue displaces healthy tissue; and infiltrating, in which healthy tissue has become diseased. Mass-effect brain tumors cause structural asymmetry by displacing healthy tissue, and may cause radiometric asymmetry in adjacent normal structures due to edema (Lefebvre, Berger, & Laugier, 1998).

For this reason, it is no surprising that the asymmetry can be utilized as one of the major indicators for the presence of brain tumors. In this study, asymmetry is determined by a novel neural network based segmentation procedure using transforms.

The constitution of the right data space is a common problem in connection with segmentation. The features that are sufficiently representative of the physical process must be searched. In the literature, it is observed that different transforms are used to extract desired information from biomedical images. Image intensities at the neighboring pixels (Dokur & Ölmez, 2000) are utilized to represent the tissues in magnetic resonance and computed tomography images. Wavelet transform (Dokur et al., 2006, Iscan et al., 2006), co-occurrence matrix (Haering & Lobo, 1999), Fourier transform (Feleppa et al., 1996), spatial gray-level dependence matrices (Llobet, Perez-Cortes, Toselli, & Juan, 2007) and Laws’ micro-texture energies (Hsiang & Sun, 2006) are used to extract tissues in ultrasound images. Wavelet transform is used for the detection of the micro-calcifications in digital mammograms (Zhang, Yoshida, Nishikawa, & Doi, 1998). The second-order statistical methods include the gray-level co-occurrence matrices (GLCM, Lefebvre et al., 1998). In GLCM method, there is an inherent problem to choose the optimal inter-pixel distance in a given situation. Sixteen features based on sizes and shapes of lesions were computed for the malignant and benign lesions of the mammography images (Retico, Delogu, Fantacci, & Kasae, 2006). The massive lesions were identified within a rectangular region of interest interactively chosen by a radiologist.

In pathological cases, the methods based on supervised or unsupervised classification integrating anatomical templates (Kaus et al., 2001, Warfield et al., 2000) have shown their robustness. Level set methods are also used for brain tumor segmentation (Ho, Bullitt, & Gerig, 2002) with some success. Mancas and Gosselin (2003) used the iterative watersheds to segment the brain tumor with a given initialization. Gibbs, Buckley, Blackband, and Horsman (1996) combined morphological process and region growing method for tumor volume determination. Cabral, White, Kim, and Effmann (1993) proposed an interactive segmentation of brain tumor based on a three-dimensional (3D) region growing. Clark et al. (1998), and Udupa and Samarasekera (1996) have introduced knowledge based techniques to make classification and segmentation more intelligent. Based on the concept of fuzzy logic, Udupa, Saha, and Lotufo (2002), and Saha and Udupa (2001) used the fuzzy clustering or the fuzzy connectedness for addressing the problem of abnormal tissue segmentation and classification. In spite of the power of those kinds of approaches, some of them need manual tracing (Udupa et al., 1997) for the initialization or a semi-supervised system (Saha & Udupa, 2001) with some manual learning.

The healthy human brain is largely symmetric across the mid-sagittal plane, recognizing that structural asymmetry may indicate disease. In previous studies, shape and volume differences between the left and right hippocampi in patients with schizophrenia (Csernansky et al., 1998, Styner and Gerig, 2001) and Alzheimer’s disease (Csernansky et al., 2000) were examined. These results have provided only a rough qualitative difference between the tumor and normal controls and that robust statistical inference should not be drawn from them. Most other work involving structural asymmetry has focused on small-scale geometric inter-hemispheric differences (Smith and Jenkinson, 1999, Thompson et al., 2001). Up to now, little attention has been paid to gross differences between the left and right brain hemispheres in patients with brain tumors.

Reviews of mid-sagittal plane (MSP) extraction methods can be found in Prima, Ourselin, and Ayache (2002). Some of the methods are based on inter-hemispheric fissure identification and symmetry criteria (Mancas & Gosselin, 2003). They exploit the extraction of symmetry lines in axial or coronal 2D slices and fitting a 3D plane to the dataset. There are methods based on cross-correlation and symmetry measures (Llobet et al., 2007, Udupa et al., 2002), but they are sensitive to asymmetry (Junck et al., 1990, Prima et al., 2002). A large deformation image warping method (Joshi, Lorenzen, Gerig, & Bullitt, 2003) is used for finding the plane of symmetry, not necessarily being the MSP. Problem of symmetry detection arises in different areas, e.g. computer vision, object recognition, shape representation, etc. Different approaches for this problem can be found in Sun and Sherrah (1997).

In the literature mentioned above, it is observed that brain tumors are detected by two different methodologies; mid-sagittal plane extraction methods, and segmentation or classification processes. In this study, these methodologies are unified to inquire the presence of asymmetry and hence, the brain tumors. First, MR head image is coarsely segmented with ISNN (incremental supervised neural network) by using feature vectors formed by continuous wavelet transform. Then, after determining the symmetry axis on this coarsely segmented image, presence of asymmetry is investigated on the segmented image by comparing the Zernike moments of the brain tissues lying on the left and right hand sides of the symmetry axis. Finally, tumorous tissues are determined and tumor is visualized apart from the healthy tissues.