Restricting the maximum boundary of region growing algorithm

Hi,

I am trying to implement a multilayer region growing algorithm for breast tumor segmentation. This is a reference of a previous question that I asked.

https://www.mathworks.com/matlabcentral/answers/781898-unable-to-select-seed-points-manually-in-region-growing-algorithm

TM25.jpg is the original image. I want to restrict the boundary of the region growing algorithm. The boundary is defined as the rectangular region in e2.jpg. I want to manually select the seed point inside the rectangular ROI. However, after implementing the multilayer region growing algorithm , I get the result shown in e1.jpg which is not correct. Any suggestions would be appreciated.

I=im2double(imread('TM25.jpg'));
I=rgb2gray(I);
imshow(I)
k = waitforbuttonpress;
point1 = get(gca,'CurrentPoint');    % button down detected
finalRect = rbbox;                   % return figure units
point2 = get(gca,'CurrentPoint');    % button up detected
point1 = point1(1,1:2);              % extract x and y
point2 = point2(1,1:2);
p1 = min(point1,point2);             % calculate locations
offset = abs(point1-point2);         % and dimensions
x = [p1(1) p1(1)+offset(1) p1(1)+offset(1) p1(1) p1(1)];
y = [p1(2) p1(2) p1(2)+offset(2) p1(2)+offset(2) p1(2)];
hold on
axis manual
plot(x,y)
[x,y]=getpts;x=round(x);y=round(y);
a=imgaussfilt(I,2);
b=adapthisteq(a);
m=regiongrowing_MLT(b,x,y,12);
m=imfill(m,'holes');
figure,imshow(m)

Answer 1

[Image Analyst]

You can call regionprops() on the regions and ask it for the perimeter. Take appropriate action if the perimeter is more than you like

props = regionprops(mask, 'Perimeter');
allPerimeters = [props.Perimeter];
if any(allPerimeters) > maxAllowablePerimeter
    % Do something...
end

[Warid Islam]

Hi Image Analyst,

Is it possible to select the boundary of the region growth interactively by using the mouse? I will be testing the algorithm on different images. The suspected tumor can be on different areas in every image. I was wondering if the user can manually select the boundary with the computer mouse.

[Image Analyst]

Yes, you can use ginput and then get the x,y from that and determine which boundary has the point that is closest to where they clicked.

boundaries = bwboundaries(mask);
[xu, yu] = ginput(1);
minDistance = inf;
for k = 1 : numel(boundaries)
    thisBoundary = boundaries{k};
    x = thisBoundary(:, 2);
    y = thisBoundary(:, 1);
    distances = sqrt((x - xu).^2 + (y - yu).^2);
    thisMinDistance = min(distances)
    % If this one is the closest, log it.
    if thisMinDistance < minDistance
        minDistance = thisMinDistance;
        selectedIndex = k;
    end
end
fprintf('You chose boundary #%d.\n', selectedIndex);
% Highlight the one they picked.
thisBoundary = boundaries{selectedIndex};
x = thisBoundary(:, 2);
y = thisBoundary(:, 1);
hold on;
plot(x, y, 'm-', 'LineWidth', 2);

They should be able to click anywhere near the perimeter or even inside it and it will identify the correct one and plot it in magenta.

[Warid Islam]

I am having a confusion about the variable mask. Are you refering to mask=imread('TM25.jpg')?

[Image Analyst]

No. Somehow you segment the image (presumably via your region growing function called regiongrowing_MLT) to create a mask (a logical/binary image). I believe it has the, unfortunately, very un-descriptive name of "m" in your code. That is probably th binary image (the mask). Is it not? What would you call the badly-named variable "m" if you were to give it a more descriptive name, which would make your code more maintainable and readable by others?

[Warid Islam]

Hi Imgae Analyst,

I tried the above algorithm, but I am coming up with the following result. It is selecting the entire image as the boundary whereas I want to select only a few portion of the image.

[Image Analyst]

You don't apply it to a screenshot of your gray scale image. Why is there that huge white surround around it anyway?

You need to apply regionprops() to your SEGMENTED image. So you need to find blobs of interest somehow and apply it to that binary image, not the grayscale image or screenshot image.

[Warid Islam]

That helps a lot. Thank you.