Inter-slice Reconstruction of MRI Image Using One Dimensional Signal Interpolation

Similar documents
Region Weighted Satellite Super-resolution Technology

IMPROVED MOTION-BASED LOCALIZED SUPER RESOLUTION TECHNIQUE USING DISCRETE WAVELET TRANSFORM FOR LOW RESOLUTION VIDEO ENHANCEMENT

SUPER RESOLUTION RECONSTRUCTION OF CARDIAC MRI USING COUPLED DICTIONARY LEARNING

Image Acquisition Systems

Investigation of Superresolution using Phase based Image Matching with Function Fitting

SUPER-RESOLUTION RECONSTRUCTION ALGORITHM FOR BASED ON COUPLED DICTIONARY LEARNING CARDIAC MRI RECONSTRUCTION

Super-Resolution Rebuilding Of Cardiac MRI Using Coupled Dictionary Analyzing

Intraoperative Prostate Tracking with Slice-to-Volume Registration in MR

COMPREHENSIVE QUALITY CONTROL OF NMR TOMOGRAPHY USING 3D PRINTED PHANTOM

A Frequency Domain Approach to Super-Resolution Imaging from Aliased Low Resolution Images

Image Reconstruction from Multiple Projections ECE 6258 Class project

BME I5000: Biomedical Imaging

Plane Wave Imaging Using Phased Array Arno Volker 1

ISSN: X Impact factor: 4.295

Basic principles of MR image analysis. Basic principles of MR image analysis. Basic principles of MR image analysis

Image Registration. Prof. Dr. Lucas Ferrari de Oliveira UFPR Informatics Department

Introduction to Image Super-resolution. Presenter: Kevin Su

Nonrigid Registration using Free-Form Deformations

Whole Body MRI Intensity Standardization

Optimized Super Resolution Reconstruction Framework for Cardiac MRI Images Perception

Computed tomography of simple objects. Related topics. Principle. Equipment TEP Beam hardening, artefacts, and algorithms

3D Surface Reconstruction of the Brain based on Level Set Method

Evaluation of Interpolation and Registration Techniques in Magnetic Resonance Image for Orthogonal Plane Super Resolution Reconstruction

An Iterative Approach for Reconstruction of Arbitrary Sparsely Sampled Magnetic Resonance Images

Machine Learning for Medical Image Analysis. A. Criminisi

WEINER FILTER AND SUB-BLOCK DECOMPOSITION BASED IMAGE RESTORATION FOR MEDICAL APPLICATIONS

PATCH BASED COUPLED DICTIONARY APPROACH FOR CARDIAC MRI IMAGES USING SR RECONSTRUCTION ALGORITHM

A Non-Linear Image Registration Scheme for Real-Time Liver Ultrasound Tracking using Normalized Gradient Fields

Spatio-Temporal Registration of Biomedical Images by Computational Methods

INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET)

MULTIMODAL MEDICAL IMAGE FUSION BASED ON HYBRID FUSION METHOD

General and Efficient Super-Resolution Method for Multi-slice MRI

An Automated Image-based Method for Multi-Leaf Collimator Positioning Verification in Intensity Modulated Radiation Therapy

Interpolation of 3D magnetic resonance data

Computational Medical Imaging Analysis

Image Guidance and Beam Level Imaging in Digital Linacs

Chapter 3 Set Redundancy in Magnetic Resonance Brain Images

Computed tomography (Item No.: P )

FOREWORD TO THE SPECIAL ISSUE ON MOTION DETECTION AND COMPENSATION

Super-Resolution. Deepesh Jain. EE 392J Digital Video Processing Stanford University Winter

Towards an Estimation of Acoustic Impedance from Multiple Ultrasound Images

Medical Images Analysis and Processing

Edge-Preserving MRI Super Resolution Using a High Frequency Regularization Technique

Classification of Subject Motion for Improved Reconstruction of Dynamic Magnetic Resonance Imaging

The University of Chicago. Center for EPR Imaging in Vivo Physiology. Image Registration. Boris Epel

RADIOMICS: potential role in the clinics and challenges

A Study of Medical Image Analysis System

REAL-TIME ADAPTIVITY IN HEAD-AND-NECK AND LUNG CANCER RADIOTHERAPY IN A GPU ENVIRONMENT

Refraction Corrected Transmission Ultrasound Computed Tomography for Application in Breast Imaging

Department of ECE, SCSVMV University, Kanchipuram

maximum likelihood estimates. The performance of

Image Registration I

APPLICATION OF RADON TRANSFORM IN CT IMAGE MATCHING Yufang Cai, Kuan Shen, Jue Wang ICT Research Center of Chongqing University, Chongqing, P.R.

Medical Image Registration by Maximization of Mutual Information

MEDICAL IMAGE ANALYSIS

A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation

coding of various parts showing different features, the possibility of rotation or of hiding covering parts of the object's surface to gain an insight

Clinical Importance. Aortic Stenosis. Aortic Regurgitation. Ultrasound vs. MRI. Carotid Artery Stenosis

Volume Rendering. Computer Animation and Visualisation Lecture 9. Taku Komura. Institute for Perception, Action & Behaviour School of Informatics

White Pixel Artifact. Caused by a noise spike during acquisition Spike in K-space <--> sinusoid in image space

MRI Physics II: Gradients, Imaging

Annales UMCS Informatica AI 1 (2003) UMCS. Registration of CT and MRI brain images. Karol Kuczyński, Paweł Mikołajczak

METRIC PLANE RECTIFICATION USING SYMMETRIC VANISHING POINTS

IMAGE RECONSTRUCTION WITH SUPER RESOLUTION

Volume 2, Issue 9, September 2014 ISSN

CHAPTER 4 RAY COMPUTATION. 4.1 Normal Computation

Marcel Worring Intelligent Sensory Information Systems

Slide 1. Technical Aspects of Quality Control in Magnetic Resonance Imaging. Slide 2. Annual Compliance Testing. of MRI Systems.

Shadow casting. What is the problem? Cone Beam Computed Tomography THE OBJECTIVES OF DIAGNOSTIC IMAGING IDEAL DIAGNOSTIC IMAGING STUDY LIMITATIONS

Edge Detection for Dental X-ray Image Segmentation using Neural Network approach

Video Mosaics for Virtual Environments, R. Szeliski. Review by: Christopher Rasmussen

2D Rigid Registration of MR Scans using the 1d Binary Projections

Tomographic Reconstruction

Abstract. 1. Introduction

Medical Image Fusion using Rayleigh Contrast Limited Adaptive Histogram Equalization and Ant Colony Edge Method

Motion Estimation for Video Coding Standards

MEDICAL EQUIPMENT: COMPUTED TOMOGRAPHY. Prof. Yasser Mostafa Kadah

WATERMARKING FOR LIGHT FIELD RENDERING 1

Advanced Visual Medicine: Techniques for Visual Exploration & Analysis

Digital Image Processing

UNIVERSITY OF SOUTHAMPTON

Segmentation of Bony Structures with Ligament Attachment Sites

Computational Medical Imaging Analysis Chapter 4: Image Visualization

Dense 3D Reconstruction. Christiano Gava

ECSE 626 Project Report Multimodality Image Registration by Maximization of Mutual Information

A fast breast nonlinear elastography reconstruction technique using the Veronda-Westman model

PET AND MRI BRAIN IMAGE FUSION USING REDUNDANT WAVELET TRANSFORM

Global Thresholding Techniques to Classify Dead Cells in Diffusion Weighted Magnetic Resonant Images

Lecture 6: Medical imaging and image-guided interventions

Enhao Gong, PhD Candidate, Electrical Engineering, Stanford University Dr. John Pauly, Professor in Electrical Engineering, Stanford University Dr.

Scientific Visualization Example exam questions with commented answers

PROSTATE CANCER DETECTION USING LABEL IMAGE CONSTRAINED MULTIATLAS SELECTION

Digital Volume Correlation for Materials Characterization

Image Interpolation using Collaborative Filtering

Functional MRI in Clinical Research and Practice Preprocessing

TUMOR DETECTION IN MRI IMAGES

Development of 3D Model-based Morphometric Method for Assessment of Human Weight-bearing Joint. Taeho Kim

Novel Iterative Back Projection Approach

Biomedical Image Analysis based on Computational Registration Methods. João Manuel R. S. Tavares

SUPER-RESOLUTION RECONSTRUCTION OF CARDIAC MRI BY USING COUPLED DICTIONARY LEARNING

Transcription:

IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.10, October 2008 351 Inter-slice Reconstruction of MRI Image Using One Dimensional Signal Interpolation C.G.Ravichandran 1, Dr.G.Ravindran 2 1 Research Scholar, Centre for Medical Electronics, Anna University, Chennai 2Chariman Information and Communication Engineering & Director, Center for Medical Electronics, College of Engineering, Anna University, Chennai. Abstract There are number of methods for creating high-quality till images from a low quality image sequences. These methods are formulated in the space domain and they require that the input be expressed in that format. Super-resolution, an open-ended research problem still hasn't made its place in Video Processing. It is still in its initial stage of research. In MRI reconstruction, based on super-resolution back-projection algorithm, the images obtained from sets of irregularly located frequency domain samples are combined into high resolution image. The superresolution replaces usually applied direct averaging of lowresolution images. Even though the super-resolution based reconstruction produces very good results in terms of quality, the cost of such reconstruction process in terms of CPU time is very high. In this paper, one dimensional signal interpolation method has been implemented for inter-slice reconstruction of 2D MRI image. The result of the one dimensional signal interpolation method for inter-slice reconstruction of MRI image is highly comparable with the existing super resolution methods in terms of speed and quality. Key words: MRI, CPU, reconstruction, super-resolution, back-projection, signal interpolation Introduction Magnetic Resonance Imaging Magnetic resonance imaging (MRI) is a noninvasive, usually painless medical test that helps physicians diagnose and treat medical conditions. MR imaging uses a powerful magnetic field, radio waves and a computer to produce detailed pictures of organs, soft tissues, bone and virtually all other internal body structures. MRI does not use ionizing radiation like x-rays. Detailed MR images allow physicians to better evaluate parts of the body and certain diseases that may not be assessed adequately with other imaging methods such as x-ray, ultrasound or computer tomography. Currently, MRI is the most sensitive imaging test of the head, particularly in the brain in routine clinical practice. MRI slices are usually thick due to hardware limitations. So a small ser of 2D slices does not give a good isotropic 3D image. This results in resolution that is high in-plane and is low in the other direction. So it is necessary to reconstruct intermediate planes/slices for better 3D reconstruction. High-quality images are assured only when the patient remains perfectly stand still while the images are being recorded. A person who is very obese may not fit into the opening of a conventional MRI machine. The presence of an implant or other metallic object often makes it difficult to obtain clear images and patient movement can have the same effect. MRI generally is not recommended for patients who have been acutely injured. This is because traction devices and life support equipment must be kept away from the area to be imaged. Furthermore, the examination takes longer than other imaging modalities and the results may not be immediately available as is often necessary in trauma situations. Also, MRI may not be as sensitive as CT in the detection of intracranial bleeding. Although there is no reason to believe that magnetic resonance imaging harms the fetus, the effects of a strong magnetic field are not well understood. For this reason pregnant women usually are advised not to have an MRI exam unless medically necessary. MRI may not always distinguish between tumor tissue and edema fluid. It cannot detect calcium present in a tumor. Detection of calcium (in tumors or other issues) is limited with MRI. MRI typically costs more and may take more time to perform than other imaging modalities. Super Resolution Super Resolution aims at reconstructing an image at a higher resolution than the set of original images from which it was derived. The core idea behind Super Resolution is to combine information contained in different images of the same scene. However, when taking images of a scene, the acquisition device often moves Manuscript received October 5, 2008 Manuscript revised October 20, 2008

352 IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.10, October 2008 which leads to a non-alignment of the different images. Therefore, it is necessary to align them perfectly to be able to combine the information. This is called as the registration process. Usually this process involves geometric component, photometric component, illumination or color balance. Once the images are registered, the Super Resolution process, estimate the high resolution image from a low resolution. This implies the construction of a forward model which describes the measured data depending on the unknown high resolution image. Once this model is defined, the next step is to construct a cost function which will measure the closeness between the measured data and the desired solution. The high resolution is then estimated minimizing the cost function. Generally, the rectifying the distortion of Image sequence using super resolution technique consists of two major steps. (1) Registration of a set of input MRI images and (2) Reconstruction of a high resolution image from them. Patrick Vandewalle et al, 2004 proposed a frequency domain technique to precisely register a set of aliased images; a high resolution image is then reconstructed using cubic interpolation. B. Marcel et al, proposed a motion estimation techniques in the frequency domain of the Fourier transform has been used for rigid-body transformation. L. Lucchese et al, 2000 defined a new method for estimating planar roto-translations that operates in the frequency domain for all the image information. Patrick Vandewalle et al, 2003 presented a simple method to generate a new high resolution image with almost twice the resolution in each dimension from a set of low resolution under sampled and shifted images. Raymond S. Wagnera et al, 2005 discussed superresolution techniques in general and review the details of one such algorithm from the literature suited to real-time application on forward-looking infrared (FLIR) images. Hayit Greenspan et al, 2002, presented a method for MRI reconstruction using super-resolution and shown improvement of spatial resolution,when spatially-selective RF pulses are used for localization. In 2D multi-slice MRI, the resolution in the slice direction is often worse than the in-plane resolution. This method improved the signal-tonoise efficiency of the data acquisition and effectively breaks the limits on slice thickness posed by the physical properties of existing MR imaging hardware. Multiple image registration and reconstruction methods have been proposed in several works. Image sequence enhancement using sub-pixel displacement was proposed by D. Keren et al,1988. A Frequency Domain Approach to Registration of Aliased Images with Application to Super-Resolution was developed by P. Vandewalle,et al, 2006. To evaluate the algorithms, few MRI image sets which were obtained from Medical research institutes are used. The performance of the algorithm in terms of time and accuracy has been measured and analyzed. In Super-resolution, motion estimation is used to estimate the pixel positions of the three images with respect to the 1st image. Note that these pixel positions would not be at integer values, but can take any real value. Once this information is calculated accurately, it is possible to project this information on a desired highresolution grid. The rotation calculation and shift calculation for the above algorithms has been implemented as follows: Rotation Calculation There are two ways that can be used to carry out rotation estimation between low resolution images. Rotate individual images at all the angles and correlate them with the first image. The angle that gives the maximum correlation is the angle of rotation between them. The method described above is computationally expensive, so a faster method is to calculate the image energy at each angle. This will give a vector containing average energy at each angle. The correlation of these energy vectors of different images will give the angle of rotation. The energy calculation could also be carried in frequency domain. A graphical representation shown below in figure 1 helps to illustrate the method clearly. Anglei = max index (correlation (I1(θ), Ii (θ))) LR image1 Energy at angle I i (θ) Shift Calculation LR image 2 Energy at angle I(θ) Figure 1 : Rotation Calculation

IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.10, October 2008 353 Once rotation angle is known between different images, shift calculation can be performed. Before calculating the shift, all the images are rotated with respect to the first image. Block matching was not used here because it is computationally expensive and need sub-pixel accuracy that is very hard to get with block matching. A much faster and reliable way to achieve this is using frequency domain method. Technically, each frequency bin should give the same shift, but that does not happen. So, a least square method is used to estimate the shift Δs. As the LR images could be aliased, only 6% of the lower frequencies were used to calculate Δs. Once we have a good estimate of shift and rotation between all the images and the reference image, we can find the pixel positions of all the Low Resolution images in the reference frame of the first image. Inter-slice Reconstruction of MRI Image Using One Dimensional Signal Interpolation The following algorithm explains the Inter Slice Reconstruction using One Dimensional Signal Interpolation. Convert the two dimensional image slices s1 and s3 into one dimensional image signal. The following equation represents the signal corresponding to the input slice s1 as y1=f(x1) The intensity values of slice s1 are mapped to some pseudo index values. The Input Slices IM202004.tif The pseudo index values of x1 will be 1, 3, 5,... n The following equation represents the signal corresponding to the input slice s3 as y3=f(x3) The intensity values of slice s3 are mapped to some pseudo index values. The pseudo index values of x3 will be 2, 4, 6, n-1, The intermediate slice s2 can be represented by the equation as y2 for known x2. Where the intensity values of the slice s2 which is proposed to be reconstructed are mapped to some pseudo index values. The pseudo index values of x2 will be 1.5, 3.5, 5.5, n The interpolation using (x1, y1),( x3, y3) and x2 will yield a new y2. Any non linear interpolation may be used by combining the two known x and y values (x1,x3 and y1,y3) and use the known (assumed) x2. we interpolate the middle layer values. Let Xk = { x1, x3 }, and Yk = { y1, y3 }, y2 can be calculated as y2= I (Xk,Yk, x2); Where I is the interpolation function. Implementation And Results The super resolution based image distortion removal and inter-slice reconstruction system has been implemented using Matlab7 on a Pentium IV, 2 GH computer. The comparative results were presented below. MRI slices available with Matlab has been used as samples. The input slices and its inter-slice output using super-resolution methods and the one dimensional signal interpolation method are shown below in the figure 2 IM202005.tif Input Slices The Reconstructed Slices

354 IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.10, October 2008 Vandewalle Kern Marcel. Reconstructed Inter Slice Using 1D Signal Interpolation Algorithm Reconstructed Inter Slice Using Superresolution Algorithms Figure 2. Inter-slice reconstructed Outputs The following table 1 shows the performance of the three super resolution algorithms and one dimensional signal interpolation method. I. Time Taken For Inter Slice Reconstruction (sec) Sl. No 1 2 3 Input slices MRISlice1.TIF MRISlice2.TIF MRISlice4.TIF MRISlice5.TIF Marcel. Kern Vandewalle 1D Signal Interpolation 90.44 96.77 149.38 1.52 91.34 97.04 150.23 1.46 92.54 98.35 148.99 1.49 II. Average Time 91.44 97.39 149.53 1.49 Table 1 : Performance in terms of Speed Size of the Slice 256x256

IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.10, October 2008 355 The following graph also shows the performance of the three implemented algorithms with input hand image of 256x256. Time Takern for inter Slice Reconstruction 160 149.5333333 Time(sec) 140 120 100 80 60 40 91.44 97.38666667 20 0 Marcel. Kern Vandewalle Algorithm 1.49 Proposed Figure 4. The Performance Graph In the above graph, the x-axis shows the algorithms and in the y-axis, the time taken for Inter Slice Reconstruction is shown in seconds. III. PSNR (db) Sl. No Input slices Marcel. Kern Vandewalle 1D Signal Interpolation 1 MRISlice1.TIF 23.85 20.15 19.98 22.95 2 MRISlice2.TIF MRISlice4.TIF 19.49 15.97 16.93 21.63 3 MRISlice5.TIF 21.34 19.45 19.01 21.91 IV. Average PSNR 21.56 18.52 18.64 22.16 The quality of reconstruction measured in terms of PSNR is shown below.

356 IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.10, October 2008 Quality of Reconstruction (PSNR) PSNR (db) 23 22 21 20 19 18 17 16 21.56 18.52333333 18.64 22.16333333 Marcel. Kern Vandewalle Algorithm Proposed Conclusion & Scope for Further Enhancements The arrived results were more significant and comparable. The implementations were made successfully on Matlab and the performances of the algorithms were analyzed with suitable metrics. Some of the important observations are: 1. Image Registration is extremely essential for successful Super-resolution. 2. If the lowresolution images have redundant data, then Superresolution does not work very well. 3. If the lowresolution images have all the information about the highresolution, i.e. they have zero redundant data among them, and then Super-resolution works perfectly. One important fact about the algorithms doesn t require tweaking of many parameters, so same code works well for a wide range of images. As far as the tests made with different images, the one dimensional signal interpolation based image reconstruction for shift and rotation estimation, performed very well in terms of speed and quality. The arrived results were significant and most promising. The performance of the algorithm makes it a real candidate for future research towards real-life implementations of the concept. REFERENCES [1] Danny Keren, Shmuel Peleg, Rafi Brada, Image Sequence Enhancement Using Sub-pixel Displacements, IEEE, 1988 [2] Luca Lucchese and Guido Maria Cortelazzo, A Noise-Robust Frequency Domain Technique for Estimating Planar Roto-Translations, IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 48, NO. 6, JUNE 2000, [3] Patrick Vandewalle, Sabine S and Martin Vetterli, A Frequency Domain Approach to Registration of Aliased Images with Application to Super-resolution, EURASIP Journal on Applied Signal Processing Volume 2006, [4] Hayit Greenspan, Sharon Peled, GalOz and Nahum Kiryati, MRI Inter-slice Reconstruction Using Super-Resolution, MICCAI 2002 [5] Chong Fan, Jianjun Zhu,Jianya Gong,Cuiling Kuang, Super-Resolution Sequence Image Reconstruction Based on Improvement Approach of Keren Registration, Proceedings of the Sixth International Conference on Intelligent Systems Design and Applications (ISDA'06), 2006 [6] Chong Fan,Jianya Gong,Jianjun Zhu,Lihua Zhang, An Improvement Approach Based On Keren Sub- Pixel Registration, ICSP2006 Proceedings,2006 [7] Patrick Vandewalle, Sabine Süsstrunk and Martin Vetterli, "A Frequency domain approach to registration of aliased images with application to super-resolution", 2004 [8] Patrick Vandewalle, Sabine Susstrunk and Martin Vetterli, "Super-resolution images reconstructed from aliased images", 2003 [9] B.Marcel, M.Briot, and R.Murrieta proposed motion estimation techniques using rigid-body transformation. [10] Raymond S. Wagnera and Donald Waagenb and Mary, "Image super-resolution for improved automatic target recognition",2005.