-
Notifications
You must be signed in to change notification settings - Fork 0
/
Image Segmentation.m
116 lines (91 loc) · 4.04 KB
/
Image Segmentation.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
Name: Sudip De
Entry Name: 2023MAS7152
Course Number: BML735
%% 4.1. Compare the tumor segmentation results with ground truth masks
% (Dice coefficient (DC) and Jaccard index (JI)) for the image segmentation algorithms:
% Ostu thresholding, region growing, K means clustering, Gaussian mixture model and Active contour.
% Load images and masks
image = dicomread('brain_tumor.dcm');
% Display original image and ground truth mask
figure;
imagesc(image);
title('Original Image');
axis off;
colormap gray;
groundTruthMask = dicomread('brain_tumor_mask.dcm');
% Display ground truth mask
figure;
imagesc(groundTruthMask);
title('Ground Truth Mask');
axis off;
colormap gray;
% Convert images to binary
threshold = graythresh(image);
imageBinary = imbinarize(image, threshold);
groundTruthMask = imbinarize(groundTruthMask);
%% Initialize variables for results
algorithms = {'Ostu thresholding', 'Region growing', 'K-means clustering', 'Gaussian mixture model', 'Active contour'};
DC = zeros(1, length(algorithms));
JI = zeros(1, length(algorithms));
% Ostu thresholding
otsuMask = imageBinary;
[DC(1), JI(1)] = computeMetrics(otsuMask, groundTruthMask);
% Region growing
regionGrowingMask = regionGrowing(image);
[DC(2), JI(2)] = computeMetrics(regionGrowingMask, groundTruthMask);
% K-means clustering
kmeansMask = kmeansClustering(image);
[DC(3), JI(3)] = computeMetrics(kmeansMask, groundTruthMask);
% Gaussian mixture model
gmmMask = gaussianMixtureModel(image);
[DC(4), JI(4)] = computeMetrics(gmmMask, groundTruthMask);
% Active contour
activeContourMask = activeContour(image);
[DC(5), JI(5)] = computeMetrics(activeContourMask, groundTruthMask);
% Display results
disp('Algorithm Comparison Results:');
disp('----------------------------------');
for i = 1:length(algorithms)
disp([algorithms{i}, ':']);
disp(['Dice coefficient (DC): ', num2str(DC(i))]);
disp(['Jaccard index (JI): ', num2str(JI(i))]);
disp('----------------------------------');
end
%% Function to compute Dice coefficient and Jaccard index
function [DC, JI] = computeMetrics(segmentedMask, groundTruthMask)
TP = sum(sum(segmentedMask & groundTruthMask));
FP = sum(sum(segmentedMask & ~groundTruthMask));
FN = sum(sum(~segmentedMask & groundTruthMask));
DC = 2*TP / (2*TP + FP + FN);
JI = TP / (TP + FP + FN);
end
%% Implement your segmentation algorithms here
% Region growing
function mask = regionGrowing(image)
mask = zeros(size(image));
end
% K-means clustering
function mask = kmeansClustering(image)
mask = zeros(size(image));
end
% Gaussian mixture model
function mask = gaussianMixtureModel(image)
mask = zeros(size(image));
end
% Active contour
function mask = activeContour(image)
mask = zeros(size(image));
end
%% 4.2. Compare the segmentation accuracies calculated in Q.1
% for different algorithms and state which algorithm(s) produced the best/worst segmentation result(s) and why?
% Which is the best algorithm (you may combine multiple algorithms to create a final algorithm)?
% The best segmentation result:
The best algorithm(s) would be the one(s) with the highest Dice coefficient (DC) and Jaccard index (JI).
The Gaussian Mixture Model or a combination of algorithms might produce the best result due to its ability to model complex distributions in the data.
% The worst segmentation result:
Otsu thresholding might produce the worst results, especially if the image intensities are not well-separated into foreground and background.
%% 4.3. How to automatically initiate active contour or region growing algorithms without providing seed points or contour by manual selection on any image?
% Active Contour:
You can initialize active contour using edge detection. The edges can serve as a starting point for the active contour algorithm.
% Region Growing:
You can use some heuristic methods like the centroid or the average intensity of the tumor region to automatically initialize the region growing algorithm without manual seed selection.