BEST代写-线上编程学术专家

Best代写-最专业靠谱代写IT | CS | 留学生作业 | 编程代写Java | Python |C/C++ | PHP | Matlab | Assignment Project Homework代写

Matlab代写 | BIM 252/EEC 205 Computational Imaging Winter 2020

Matlab代写 | BIM 252/EEC 205 Computational Imaging Winter 2020

这个Project是用Matlab完成图像的模型重建

BIM 252/EEC 205 Computational Imaging Winter 2020
PROJECT
The focus of this project is on reconstruction methods. Therefore, we use simulated data to
avoid any requirement of imaging modality specific knowledge (so the competition is fair to
everyone). Specifically, the data were generated by the following Matlab codes:
load train_phantom % get the phantom image variable x
angle = [0:10:179];
ybar = radon(x,angle);
% add Poisson noise
y = poissrnd(ybar);
The following codes perform reconstruction using iradon and calculate the RMSE, which will be
used as the baseline value (see performance score calculation below).
% reconstruct using iradon
fbp = iradon(y,angle,’linear’,’ram-lak’,1,115);
% now compute RMSE
rmse = sqrt(mean((x(:)-fbp(:)).^2))
The following code can be used to generate a projection matrix H to be used in a model-based
iterative reconstruction method:
H = gen_sys_N(115, angle);
There are four data files: two for training and two for testing
train_phantom.mat contains a brain phantom image ‘x’ and its projection ‘y’
train_ct.mat contains a CT image ‘x’ and its projection ‘y’.
These two files are training data with ground truth images (shown in figure below) so
that you can use them to tune your reconstruction methods.
For the deep learning method, you will need to get training data by yourself. You can
use the phantom() function in matlab with random parameters or use online resources.
test_phantom.mat
test_ct.mat
These two files are testing data that will be used to evaluated the performance of your
methods. They contain only the noisy projection ‘y’ without the ground truth image. You will
submit your reconstructed images of the two testing data in ONE .mat file. The RMSE will be
calculated for you. The .mat file should contains 4 variables named as follows:
recon_phantom_mb : model-based reconstruction of test phantom
recon_phantom_dl : deep-learning reconstruction of test phantom
recon_ct_mb : model-based reconstruction of test CT image
recon_ct_dl : deep-learning reconstruction of test CT image
Each variable should be a 115×115 array. Please save them in a mat file with filename ‘{your last
name}_recon.mat’ (without {}) and upload it to Canvas.
It is very important that you following the above naming instructions so that we can calculate
your RMSE performance score which counts 50% of your project grade!
Figure. The brain phantom image (left) and the real CT image (right). The images used in
the project are downsampled to 115×115 to reduce computation time.
Grading
The project will be graded based on your RMSE performance on the testing data (50%) and
written report (50%).
The RMSE performance score is calculated based on the reduction of RMSE over the baseline
method, which is the reconstruction by iradon.m in MATLAB. The method achieves the lowest
RMSE will receive a full score of 100, and score of other methods will be calculated by
performance score = ���� ���� − ����6789:;
������ ���� − ����6789:;
× 100
The performance score will be calculated for every test dataset and reconstruction method
combination (4 combinations in total) separately and the average will be used as the final score.
Please note that only the RMSE of the testing data will be considered in the calculation of
performance score.
The written report should include a method section, a result section, and a discussion section.
The method section should describe the model-based reconstruction algorithm and deeplearning
based reconstruction methods and the rationale of your choice, and any procedure
that you used to optimize the performance.
The result section should include at least your (i) reconstructed images of the training data, (ii)
reconstructed image of the testing data, and (iii) the RMSE of the training data reconstruction,
for BOTH model-based reconstruction and deep-learning based reconstruction. You can also
include any intermediate results that you used to fine tune your method.
The discussion section should compare the two methods that you studied, and discuss their
respective pros and cons, as well as any limitations of your study.

bestdaixie