Icbme 2011

Icbme 2011

• 1.
  IMAGE FUSION AND PSEUDO scale, multi-spectral or multi-temporal COLORING IN MRI ANALYSIS remote sensing data and generating a new data with higher information content‟. The Project Abstract: Providing a simple, single main objectives of image fusion are modality Image fusion and pseudo-coloring improved image reliability (by redundant algorithm for better visual interpretation of information) and also improved image multiple MRI scan images of the same capability (by complementary information). section of the brain. The project is also Ideally, the method used to merge data sets compared with the ISODATA model for with high spatial and high spectral resolution better visual perception of multiple MRI should not distort the spectral characteristics scan images of the same section. The of the high-spectral resolution data. constraints of both methods are also discussed. The paper also distinguishes II. CATEGORIZATION OF IMAGE between single modality and multiple FUSION ALGORITHMS modality based image fusion in MRI while citing examples of each. The algorithms available for image fusion, operate on a pixel-level, feature-level and decision-level. We concentrate on the pixel- based fusion which is performed at the level I.INTRODUCTION of spectral radiance values and offers The purpose of an image fusion process is to minimum of original spectral information. combine a number of multimodal or This technique requires the input images to multispectral images into a final entity that be registered with high accuracy of less than comprises the maximum possible half a pixel, since incorrect registration can information, which is present in the source cause artificial colors in features of data, images. The source images often exhibit a thereby leading to falsifying of high degree of correlation since the same interpretation. area is covered in different regions of the Image fusion techniques can also be electromagnetic spectrum or with categorized into three types, color-related, complementary imaging technologies. Thus, numerical/ statistical-related and combined the same information can be found in more approaches. All color-related techniques than one of the source images and is employ slicing of original data into their described as overlapping information. respective layers, which can be basic RGB, The additional information of the human perceived IHS, HSV or more panchromatic band in combination with the scientific luminance–chrominance. This is multi-spectral bands, allows the retrieval of followed by substitution by a high resolution maximum image information from the given image in place of one of these channels and image data set. According to Pohl, Van a back-transformation of this combination Genderen and Wald, „Image fusion is a into the original RGB domain. process of performing the alliance of multi-
• 2.
  III. REVIEW OFIMAGE FUSION cluster in a feature space the axes of which TECHNIQUES CURRENTLY IN USE IN represent the signal intensity of that tissue MR IMAGING on MR images of that type. These clusters are then represented by different shades of There are several kinds of Image fusion gray and pseudo-colored using various techniques in MRI interpretation at the color-maps until a suitable colormap is moment. One such technique which applies found for the fused image. multimodal image fusion is the image fusion of an MRI scan and a CT scan for treatment planning in Tumor treatment. Treatment IV. CHANNEL BASED IMAGE FUSION planning based on fused CT and MRI data AND PSEUDOCOLORING enables better definition of target volume An MRI is defined by its tissue selectivity and risk structures as compared to treatment for contrast, and physicians generally derive planning based on CT alone. Here, the diagnostic information from an MRI image fusion technique only fuses a single because a particular tissue is displayed with MRI scan based image with a CT scan a different contrast. However, each of these image and is useful for registration of the MRI scans having different diagnostically parts(to view the critical organs while useful data can be differently identified viewing the bone-related information). The using machine parameters such as term „modalities‟ here refers to either of Repetition Time, Echo Time and Inversion „CT‟, „MRI‟, „PET‟ and „SPECT‟. The Time. Most of these parameters are Image fusion techniques presently in use available in the DICOM header file which is combine images of different modalities to filled in by the machine. There are various form a fused image. This is often mentioned commonly used MRI scans, namely, T1 in publications as the 3TP method. weighted image scan, T2 weighted image In terms of single modality based image scan, T1 FLAIR weighted image scan, Post fusion, a particular technique has been Contrast T1 weighted image scan. identified as well as published. In this We can combine these different scans for publication, a multi-parametric MR image better visual perception of data by Image set was analyzed with the iterative self- fusion with channel based coloring of organizing data (ISODATA) technique and individual MRI scans. For example, for it consisted of T1-weighted images, fat- three MRI scans T2 weighted image, T1 suppressed T2-weighted images, and three- FLAIR weighted image and Post Contrast dimensional fat-suppressed T1- weighted T1 weighted scan we get the fused and images which were acquired before and channel colored image as: during contrast material enhancement (see MR Imaging).These imaging sequences constitute the conventional breast MR = imaging examination, and they were selected for ISODATA analysis because These parameters were obtained by each sequence provides different contrast to checking each of the physician identified disclose different tissue types. images and seeking the parameters from the If we assume that each tissue type has DICOM header file. characteristic signal intensity on each MR image type, then each tissue type will form a
• 3.
  Parameter T1 T2 T2 PCT1 contrast between a fat tissue and a water FLAIR FLAIR tissue while a FLAIR sequence image forms Echo >60 60 <30 <30 a set of elements while nullifying the fluid Time data in the image. This can be thought of Inversion More Less 500- Less two sets of elements with a common factor. Time than than 1000 than 1000 1000 1000 Identifying both these sets and their Repetition >3000 > 1500- 1500- common elements must be done for Time 3000 3000 3000 diagnosis of the problem. This can be further Contrast - - - „IV‟ simplified by fusing the images and giving Info each of these series a particular color channel. The physician can then diagnose the problem by identifying regions of Are the parameters for a „GE‟ MRI scan problem occurrence in each of the machine of 1.5 T field strength. underlying colors. This method, helps V.MRI SEQUENCES AND THEIR because, it provides greater amount of INFORMATION information in the image as well as better visual interpretation of the image. Each of these series provides a different form of information to the physician. For T2 weighted image (also referred to example, Fluid attenuated inversion as T2WI) is one of the basic pulse recovery (FLAIR) is a pulse sequence sequences in MRI and demonstrates the differences in the T2 relaxation time of an inversion recovery technique that nulls tissues. The T2WI relies upon the transverse fluids. It can be used in brain imaging to relaxation of the net magnetization vector suppress cerebrospinal fluid (CSF) effects (NMV). T2 weighting tend to have on the image, so as to bring out the peri- long TE and TR times. In a T2 weighted ventricular hyper-intense lesions, such image, the fat portion of a tissue appears as multiple sclerosis (MS) plaques. Its intermediate bright whereas the water usefulness is different from a T1 weighted portion appears very bright. image. T1 weighted image (also referred to So if the doctor wants to identify between as T1WI) is one of the basic pulse white matter and gray matter in the brain. sequences in MRI and demonstrates the He cannot do that purely on the basis of a T1 differences in the T1 relaxation time of weighted image. He needs to look into both tissues. T1WI relies upon the longitudinal the T1 weighted image and the T2 weighted relaxation of the Net Magnetization Vector. image. The white matter is wrapped in a Fat has a large longitudinal and transverse fatty layer called myelin, which insulates the magnetization vector and hence appears axons and allows them to conduct signals bright on a T1 weighted image. On the other quickly, much like rubber insulation does hand, water appears to have less longitudinal for electrical wires. The type of fat in myelin magnetization prior to the RF pulse. Thus, makes it look white, so myelin-dense white water has low signals and appears dark. matter takes on a white hue as well. Because Therefore the T1 weighted image shows a gray cells are not surrounded by white
• 4.
  myelin, they takeon the natural grayish automated tumor identification and color of the neurons and glial cells. Now classification. This approach may enable the suppose that the T1 weighted MRI was identification of specific tissue signatures fused with the T2 weighted MRI. We could characterisic of benign versus malignant from this fused image directly distinguish tumors. between gray matter and white matter which was previously a function of a T1 weighted REFRENCES image alone. Also the function of a T2 [1] J.H. Jang and J.B. Ra,” Pseudo Color Image weighted image could be seen as well. fusion based on Intensity-Hue-Saturation Color (Detecting bleeds, swellings at the same Space”, IEEE Conf. on Multisensor Fusion and Integration for Intelligent systems, TE 4-3. point of time.) Passing each of these images through a specific color channel (Red, Blue [2] A. Toet, “Natural color mapping for multiband or Green) helps in fusing the three grayscale night vision imagery,” Information Fusion, vol. 4, pp. 155-166, 2003. images without much loss of information. [3] R. C. Gonzalez and R. E. Woods, Digital Image Here, in each image, the parts of the image Processing, Prentice-Hall, 2002. which are bright in a T1 weighted image [4] J. H. Jang, Y. S. Kim, and J. B. Ra, “Image will appear slightly red in the pseudo- enhancement in multi-resolution multi-sensor colored fused image. The parts of the image fusion,” Proc. IEEE AVSS, pp. 289-294, Sep. 2007. which are bright in all the three images will [5] Nargess Memaradeghi,”A Fast implementation of appear as shades of gray, while those which the ISODATA clustering Algorithm”, Itnl.J. are bright in two images will appear as a Computing and Geometry, 2006. combination of those colors, while patches [6] Tou J, Gonzales R. “Pattern recognition which are bright in a single image will principles. Reading, Mass: Addison-Wesley, appear bright in the input channel alone. 1974. [7] Michael A. Jacobs, PhD Peter B. Barker, D Phil, This method thus provides more information David A. Bluemke, et al, “ Benign and Malignant as well as easier visual interpretation of the Breast Lesions: Diagnosis with Multi-parametric MR boundaries. Also the edges of the tumors can Imaging”, Radiology 2003; 229:225–232 be more easily detected in the combined [8] Thorsen Twellmann, Oliver Lichte, et al ,”An image by applying edge detection Adaptive extended color scale for comparison of pseudo-coloring techniques used for DCE-MRI algorithms to each of the individual images Data”,Applied Neuroinformatics group, Department first followed by the same image fusion. of Radiology, University of Munich. Some sample images are shown below with their processed equivalent images. CONCLUSION The proposed algorithm will make it easier for doctors to make informed diagnosis based on MRI scan information for
• 5.
  (a) Post contrastT1 weighted image (b) T1 FLAIR weighted image (c) T2 FLAIR weighted image (d) Fused and Pseudo-colored image