CT and MRI Image Fusion based on Wavelet Transform and Neuro-Fuzzy concepts with quantitative analysis S Rajkumal1, Puja Bardhan#2, Satya Kumar Akkireddl3, Chirag Munshi#4 "School a/Computing Science and Engineering VI T University, Vellore, India
[email protected]
Abstract-Medical image fusion is the process of combining two different modality images into single image.
This resultant
image is helpful in medical field for efficient disease diagnoses, retrieval
of
images,
undergo
surgery
treatment,
tumor
identification etc. This single image features cannot be obtained from the single modality medical images and it can be resolved by image fusion. This paper has proposed two fusion techniques Iterative
Neuro-Fuzzy
Approach
(INFA),
Lifting
Wavelet
Transform and Neuro-Fuzzy Approach (LWT-NFA). By using proposed techniques, Computed Tomography (CT) and Magnetic Resonance Image (MRI) images are fused and it has been compared
with
existing
technique
using
quantitative
and
qualitative measures. From the implementation observed that INFA algorithm provides clear image information based on the measures.
Keywords- Medical Image Fusion, Wavelet Transform, Fuzzy, Neuro-Fuzzy, Quantitative Analysis
I. INTRODUCTION
In real world, now-a-days the usage of medical image plays an important role in hospitals. These medical images such as Computed Tomography (CT), Magnetic Resonance Image (MRI), Positron Emission Tomography (PET) and Single photon emission computed tomography (SPECT) are captured using different sensor scan machines. CT identifies the bone structure, MRI image gives that the soft tissue information; PET and SPECT provide human body functionality information. As per the radiologist consent these images cannot provide clear information image for disease diagnosis and treatment planning etc. So, different modality complementary information for efficient disease diagnosis is required. This information is then passed through the multimodality medical image fusion techniques. To this point, extensive research has been conducted on many fusion algorithms for medical images. Depending on the fusion rule, the medical image fusion scheme is classified into three categories: pixel level, feature level and decision level. Medical image fusions regularly utilize the pixel level fusion techniques. At pixel level fusion famous algorithms include Average Method [1], Maximum Selection Method [I], Principle Component Analysis [2], Laplacian Pyramid Method
[3]. As image features are sensitive to human visual system they exist in different scales, multi-resolution analysis which is more suitable for image fusion. The wavelet-based methods such as Discrete Wavelet Transform (DWT) [4], Contourlet Transform [5], and Redundancy Discrete Wavelet Transform (RDWT) [6] provide solution for multi-resolution. The wavelet transform is used to isolate the discontinuities at object edges but fails to detect the smoothness along the edges. To resolve this problem in this paper two fusion techniques have been proposed viz. Iterative Neuro-Fuzzy Approach (INFA), Lifting Wavelet Transform and Neuro-Fuzzy Approach (LWT-NFA) and the resultant of this method is compared with existing technique Discrete Wavelet Transform (DWT), average method using subjective and objective measures such as Normalized Correlation Coefficient (NCC), Entropy (EN) [6], and Structural Similarity Index (SSIM). The rest of the sections are organized in this paper as follows: In section-II, the system design is briefly discussed; section-III deals with experimental results and performance evaluation of fusion techniques based on the quantitative metrics; section-IV, conclusion is summarized.
II. SYSTEM DESIGN
In this system, initially two different types of registered modality images CT (anatomical) and MRI (pathological) are considered as input. Then to get more information on input images, input images are fused with fusion techniques Iterative Neuro-Fuzzy Approach (INFA), Lifting Wavelet Transform and Neuro-Fuzzy Approach (LWT-NFA), Discrete Wavelet Transform (DWT) and Average Method. Finally, to validate the fusion techniques quantitative measures have been used. The overall system Architecture is shown in Fig. 1.
A. Iterative Neuro-Fuzzy Approach Real time application and quality of the images are two important considerations for the image to be used in industrial applications [7]. Real time image requires lesser image data processing and faster computing, so it requires fast computing platform as well as less complex algorithm. But for critical applications like medical imaging or satellite imaging or
automatic target guidance system, information content need to be reliable and should reflect various aspects of the image. Iterative image fusion technique is very useful in medical imaging and other areas where quality of the image is more important than the real time application. Once we get a fused image from input images, we can further use the same image for fusion with the one or the other input images to get a better image quality depending on the application. It can be further extended in the applications of the method by using it for fusing medical and landmine images. After fusing the input images, we can further fuse it with one or both of the input images to get a better quality image. When two or more than two images are given as inputs to the neuro fuzzy system they have equal share pixel-wise in final fused output. But in the iterative approach, some images are prioritized which are fused more than once for the final output image. Suppose we have three image sources where one is a normal image, the second is IR image and the third is image from some other source and in the final image, we want to have more effect from IR image. Then we fuse the new output image with the second IR input image once again and the process is repeated if required. In the algorithm, we have adopted a general process in which we have taken n images, and images have been given priority with some index for the priority. Index of priority for an image decides how many times an image has to be fused for the final output image. From practical point of view, we should get the better image at two to three iterations, when taken into consideration the time taken to get the fused image. 1) Algorithm: Step I: n images M),M2 . . . . Mn of equal size are read with priority index t),t2 .. tn where t1>t2 . . . . tn> I respectively. Priority index tn suggests that the image Mn should be fused "n" times for the final output image. Step2: Training data and Neuro Fuzzy structure is made using "fismat" and the "anfis" command with specified member ship function type and number. Step3: Image is fused after forming the check data of input image colunms Step4: Step 2 is repeated and the new neuro fuzzy structure formed is used for further fusion process. Step5: Depending upon the value of 'n' in 1u the image is fused 1u-l times and for each fusion �1 is incremented by 1. Step6: The fusion process is continued with two inputs in which the first input is the output image colunm and second is the required input satisfying the condition 1;