Elsevier

Image and Vision Computing

Volume 25, Issue 2, February 2007, Pages 164-171
Image and Vision Computing

A framework of fuzzy information fusion for the segmentation of brain tumor tissues on MR images,☆☆

https://doi.org/10.1016/j.imavis.2006.01.025Get rights and content

Abstract

A framework of fuzzy information fusion is proposed in this paper to automatically segment tumor areas of human brain from multispectral magnetic resonance imaging (MRI) such as T1-weighted, T2-weighted and proton density (PD) images. A priori knowledge about tumors described by radiology experts for different types of MRI are very helpful to guide a automatic and a precise segmentation. However, the terminology used by radiology experts are variable in term of image signal. In order to benefit of these descriptions, we propose to modellize them by fuzzy models. One fuzzy model is built for one type of MRI sequence. The segmentation is finally based on a fusion of different fuzzy information obtained from different types of MRI images. Our algorithm consists of four stages: the registration of multispectral MR images, the creation of fuzzy models describing the characteristics of tumor, the fusion based on fuzzy fusion operators and the adjustment by fuzzy region growing based on fuzzy connecting. The comparison between the obtained results and the hand-tracings of a radiology expert shows that the proposed algorithm is efficient. An average probability of correct detection 96% and an average probability of false detection 5% are obtained through studies of four patients.

Introduction

Accurate and robust brain tissue segmentation from magnetic resonance imaging (MRI) is a very important issue in many applications of medical image analysis for quantitative studies and particularly in the study of some brain disorders. One example is to analyze and estimate quantitatively the growth process of brain tumors, and to evaluate effects of some pharmaceutical treatments in clinic [1], [2]. A brain tumor segmentation consists of separating the different tumor tissues such as solid tumor, edema and necrosis from the normal brain tissues such as gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF). Since MRI can provide various information about brain tissues from a variety of excitation sequences, it is a powerful tool in clinic to improve diagnosis. The different types of MR images obtained from the different excitation sequences, also called multispectral images, can provide different image intensity information for a given anatomical region and subject. Since a tumor consists of different biological tissues, one type of MRI cannot give complete information about abnormal tissues. Therefore, radiology experts always combine multispectral MRI information of one patient to take a decision on the location, extension, prognosis and diagnosis of the tumors. At least three types of MRI, generally T1-weighted (T1), T2-weighted (T2) and the proton density (PD) are used as the routine sequences in the clinic. These three types are also used in our study. It is very easy to extent our method by adding other types of MRI.

As we know, manual tracing by an expert of a tumor in 3D for all types of MR modalities involved in studies is not only exceedingly time consuming, but also exhausting for experts, which can lead to human errors. Therefore, an automatic segmentation is necessary. Numerous studies of brain segmentation have been carried out and are reported in the literature. The methods based on elastic registration using elastic matching techniques, or deformable models [3], [4] have proven to be reliable and efficient for small and local shape changes [5], especially for normal tissue segmentation. The methods based on statistical models, such as Gaussian intensity models [6], explicit models [7], Markov random field models [8] work well in the case of normal tissue segmentation, but not for abnormal tissues. In pathological cases, the methods based on supervised or unsupervised classification integrating anatomical templates [1], [9] have shown their robustness. Level set methods are also used for brain tumor segmentation [10] with some success. Mancas and Gosselin [11] used the Iterative watersheds to segment the brain tumor with a given initialization. Gibbs et al. [12] combine morphological and region growing for tumor volume determination. Cabral et al. [13] proposed an interactive segmentation of brain tumor based on a three-dimensions (3D) region growing. Clark et al. [14] and Udupa and Samarasekera [15] have introduced knowledge-based techniques to make classification and segmentation more intelligent. Based on the concept of fuzzy logic, Udupa et al. [16] and Saha and Udupa [17] used the fuzzy clustering or the fuzzy connectedness for addressing the problem of abnormal tissue segmentation and classification. In spite of the power of these kind of approaches, some of them need manual tracing [18] for the initialization or a semi-supervised system [17] with some manual learning.

In conclusion, full automatic segmentation of tumor tissues is still a difficult problem for two key reasons: one is that there is a large number of tumor types which differ greatly in size, shape, location, tissue composition and tissue homogeneity [9]. In some cases, their border with normal tissues cannot be very well defined on the images, therefore they are even difficult for radiology experts to delineate. The other one is the consequence of the partial volume effects (one pixel may be composed of multiple tissue types) and the noise due to the MRI acquisition system. Taking into account these problems, a knowledge-based technique and a fuzzy segmentation are adapted for coping with these difficulties. This technique can offer a possibility value of belonging to the brain tumor for each voxel.

We present in this paper, a fuzzy information fusion framework for the automatic segmentation of tumor tissues of human brain from multiple MR image sequences. This framework consists of the registration of multispectral images, the creation of fuzzy models (based on a priori knowledge), fuzzy feature fusion and an adjustment (finesse) by fuzzy region growing. The paper is organized as follows. First, an overview of the proposed method is presented in Section 2. Then Section 3 shows some fuzzy modeling methods for establishing mathematical relationships between the knowledge of tumor tissues provided by radiology experts and the multispectral image signals. Some fuzzy combination operators are suggested in Section 4 to fuse the multispectral information extracted by the fuzzy models to obtain a coarse segmentation. In Section 5, a fine adjustment from the coarse segmentation, based on fuzzy region growing, is introduced. Some experiment results using three routine MRI sequences T1, T2 and PD feature images are shown in Section 6. We set out conclusion in Section 7.

Section snippets

Overview of the fuzzy segmentation procedure

Let us first describe in more detail the overview of our method, which is based on knowledge of experts and fuzzy concept as we discussed in Section 1. According to the descriptions of radiology experts, a glial tumor is a very bright region, which may be brighter than GM in T2 image and in PD images. But it is dark in T1, darker than GM, but often brighter than CSF. These concise descriptions are very fuzzy in term of image signals, but very helpful for us to know about relative signals of

Creation of fuzzy models

A fuzzy set has been defined as a collection of some objects with membership degrees [21]. A membership function represents a mapping of the elements of a universe of discourse to the unit interval [0, 1] in order to determine the degree to which each object is compatible with distinctive features to collect. A membership function, considered as a fuzzy model, is used to obtain a corresponding fuzzy set. Pedrycz et al. [21] have given an exhaustive analysis for experimental methods to determine

Fuzzy feature fusion

As tissues within a tumor are not homogeneous, they are more or less observed on the three types of MRI. Therefore the use of the three types of images as the advantage of given more complete information about the tumor. That is why we use information fusion. The obtained fuzzy models, μFrT1, μFrT2, and μFrPD, allow us to transform the three types of MRI volumes into three fuzzy feature spaces, noted as FrT1, FrT2, and FrPD respectively, whose values represent their possibilities belonging to

Fuzzy region growing

Region growing is one of the region-based segmentation methods in the domain of image processing. Region homogeneity is usually used as the main growing criterion, and can be based on gray level, color, texture, shape and model when using semantic information.

We propose a fuzzy region growing algorithm to improve the FμI obtained after the fusion step. It is a region-connection process based on the similarity of membership degrees and the neighborhood positions of the regions in 3D space. FμI

Evaluation and results

Data from 4 patients with glial cerebral tumor were processed in this paper. MRI images of each of the four patients are acquired on a 1.5 T GE (General Electric Co.) apparatus using an axial 3D IR (Inversion Recuperation) T1-weighted sequence (TI/TR/TE: 600/10/2), an axial FSE (Fast Spin Echo) T2-weighted (TR/TE: 3500/102) and an axial FSE PD-weighted sequence(TR/TE: 3500/11). The total number of slices for T1 image is 124, with a 094×0.94×1.5 mm3 voxel size, 20 slices for T2 and PD images, with

Conclusion

We have presented in this paper a fuzzy segmentation method based on knowledge fuzzy models and fuzzy fusion, in order to automatically segment tumor areas of human brain from multispectral MR images. Three fuzzy models are introduced to represent tumor tissue features for different MR image sequences. They allow us to create corresponding fuzzy feature spaces of tumor. All the t-norm or fuzzy intersection operators can be used as fusion operators for these fuzzy features. The geometric mean is

References (28)

  • V. Barra et al.

    Automatic segmentation of subcortical brain structures in MR images using information fusion

    IEEE Trans. Med. Imaging

    (2001)
  • M. Styner et al.

    Parametric estimate of intensity inhomogeneities applied to MRI

    IEEE Trans. Med. Imaging

    (2000)
  • K.V. Leemput et al.

    A unifying framework for partial volume segmentation of brain MR images

    IEEE Trans. Med. Imaging

    (2003)
  • Sean Ho et al.

    Level set evolution with region competition: automatic 3-D segmentation of brain tumors

    Int. Conf. Patt. Recog.

    (2002)
  • Cited by (151)

    • Study of image fusion optimization techniques for medical applications

      2022, International Journal of Cognitive Computing in Engineering
    • Medical image fusion methods: Review and application in cardiac diagnosis

      2021, Image Processing for Automated Diagnosis of Cardiac Diseases
    View all citing articles on Scopus

    NSFC60372023 supported by the National Natural Science Foundation of China.

    ☆☆

    ITIC project supported by national and regional French funds, and FEDER funds.

    View full text