Вы находитесь на странице: 1из 29

Ray-Theory In Tomography

- Abhishek Ranjan - Atrij Singhal - Mayank Jain


Under the guidance ofProf. Prabhat Munshi

Introduction
Tomographic imaging consists of assuming that the cross section consists of an array of unknowns, and then setting up algebraic equations for the unknowns in terms of the measured projection data wherein we use Algebraic Reconstruction Methods.

ART (Algebraic Reconstruction Techniques)


We consider a square-grid on an image f(x,y). In each cell the function f(x,y) in constant. A line integral is called a raysum( ).

-weighting factor: for this cell =


- for square grid intercepted by ray. - is the width of the cell.

The relationship between s and smay be expressed as:-

- M is the total number of rays. - N is the total of cell in the grid.

Method of Projections

- Each equation represents a hyper plane. - If unique solution exists, intersection of all the hyperplanes gives us the solution. - If M and N are small, we can use conventional methods to invert the system of equations. - For large values M and N

We first make an initial guess 0 . 0 is projected of the first hyperplane giving 1 and so on we get 0 , 1 . . The process can be mathematically described as:-

= (1 , 2 , 3 )

PROBLEM STATEMENT 1
To calculate the weight matrix for a 3x3 image matrix To reconstruct the image using the inverse of weight matrix To use algebraic reconstruction technique to reconstruct the image

Algorithm for reconstructing 3x3 image


We assumed the image to be 3X3 matrix. Each cell is given a value between 0-1 on a grayscale. The image on the left represents our choice of origin.

Constructing a weight matrix for 3^2 rays for a image matrix of 3x3 Find the intersecting points with the 8 lines of the image matrix. Filter the points which lie within the grid. Arrange the points in increasing y-coordinate. Find the distance between successive points and feed them into the weight matrix. Repeat the steps for other rays. Reconstruction of image from weight matrix and column matrix of ray sums. We used the ART algorithm for reconstruction. First start by assuming column matrix f to be null. Then run iterative ART algorithm.

Original Image:-

3X3 Algebraic Reconstruction Matlab Code

RESULTS
Image matrix0.2000 0.5000 0.5000 1.0000 0.0000 0.7000 Image reconstructed art:0.9000 0.4000 0.1000

Weight matrix-

Final reconstructed image using inverse of weight matrix:-

Final reconstructed image using ART Algorithm:-

(after 100 iterations)

Original Image

Error analysis
Rms difference between the pixel vales of original and the reconstructed image: Inverse determinant method - 7.1165e+015 Artificial reconstruction method - 0.0383

Conclusion
Inverse determinant method is not effective because finding the inverse of sparse matrix is not efficient in MATLAB. Art method produces more accurate results, with an error of 3.83% for 100 iterations. This error can be further reduced by increasing the no. of iterations.

PROBLEM STATEMENT 2
To calculate the weight matrix for an 64x64 image matrix and subsequently use ART to reconstruct the image matrix.

We constructed a 64x64 SHEPP-LOGAN represented as column matrix F(n) of order 64^2. We assumed the image to a 64x64 grid shown below A emitter which can emit 64 parallel rays is considered in 64 different position separated by pi/64 angle Then the weight-matrix W(i,j) for this geometry is constructed Then reconstruct the image using ART algorithm

Algorithm for reconstructing 64X64 image.


Constructing the weight matrix for 64^2 rays for a image grid of 64x64
Find the first two vertical lines which the ray intersects. Find the two intersecting points (x1,y1) and (x2,y2). Find the no. of horizontal lines between these two points and divide the distance between the two points in subsequent ratios. Feed the consequent distance for each pixel in the weight matrix. Repeat the steps for next lines until the end of the grid is reached. Repeat the steps for all the 4096 rays to obtain the weight matrix.

Reconstruction of image from weight matrix and column matrix of ray sums.
We used the ART algorithm for reconstruction

First start by assuming column matrix f to be null Then run iterative ART algorithm

Original Image

Result Image

(After 5000 iterations)

64x64 SHEPP-LOGAN MATLAB Code for the problem

Problems Faced
The algorithm used to reconstruct the 3X3 matrix could not be extended to reconstruct higher order images The time taken by the code to execute took around 1 hour. Weight matrix produced had an error of 10.33% compared to the matrix formed in the new algorithm.

Weight constructed for higher order matrix(i.e. 64X64 in our case) takes up large physical memory which slows down the code execution, this problem was overcome by evaluating each row of weight matrix and simultaneously performing the reconstruction algorithm.

PROBLEM STATEMENT 3
Reconstruction using Multiplicative Algebraic Reconstruction(MART) technique for 4X4 and 64X64 matrix.

Algorithm used for MART


The algorithm used to construct the weight matrix was the same as the algorithm used to reconstruct using ART. The iterative step for mart is-

X is assumed pixel values k is the current iteration number b is the ray sum A is weight Matrix

When the error factor(dx) was zero then that particular iteration was ignored. If the pixel value was greater than 1, it was set back to zero and if it was less than 0 then it again it was set back to zero.

Results for 4X4 matrix using MART


Original Image Final Reconstructed Image

Error analysis-

Results for 64X64 matrix using MART


Original Image Final Reconstructed Image

(after 20 iterations)

(after 10 iterations)

Conclusions
Convergence of assumed solution to original solution is achieved quicker in MART as compared to ART. The amount of noise is significantly reduced as compared to the final reconstructed image in case of ART.

Errors in both ART and MART are within acceptable limits and almost equal.

PROBLEM STATEMENT 4
Using Householder method for finding the inverse of a sparse symmetric matrix.

Algorithm for Householder method

Results
Original symmetric matrix1 -1 2 2 -1 2 1 -1 2 1 3 2 2 -1 2 1 Inverted matrix-

1.0000 3.0000 -0.0000 -0.0000

3.0000 3.7778 0.7778 0.1111

-0.0000 0.7778 2.7778 1.1111

-0.0000 0.1111 1.1111 -0.5556

---THANK YOU---

Вам также может понравиться