How do I detect lines in 3D Volume?

Hi,

I'm trying to find a method to detect these lines (lung fissures) within a 3D volume. This is an image mask created after a Hessian Matrix analysis (so i have eigen values and vectors, as well as the original CT volume). There's so much noise and I'm not sure what path to take!

Any ideas?

Thanks.

Answer 1

[Joaquin Quintana]

Hello -I pulled the image you have here and used regionprops to find the lines you have pointed out. I was unsure if it was just the two highlighted lines or you wanted others as well so I just worked on the two explicitly pointed out.

I see you want to do this on volume data and this can be done using regionprops3and a similar method but it will be slightly different and may need some adjustments. If you are going to try and batch process volume data, I would use other parameters which are unique to the fissure lines to improve the segmentation (see regionprops3 properties). Here I simply used the area property but could have used the MajorAxisLength as well to make it more robust.

Identified lines

Identified lines overlaid original image

Code used:

%read image

img = imread('image.png');

%convert RGB image to grayscale then to binary and invert

BW_invert = ~imbinarize(rgb2gray(img));

%erode binary image

se = strel('line',2,90);

erodedBW = imerode(BW_invert,se);

imshow(erodedBW);

%find connected components

cc = bwconncomp(erodedBW);

%get stats from regionprops

stats = regionprops(cc,'Area');

%keep only indexes with areas which are related to lines of interest

idx = find( ([stats.Area] > 119) & ([stats.Area] < 150) );

%use ismember to keep the indexes which are related to the lines

lines = ismember(labelmatrix(cc),idx);

%dilate data back to orignal size

lines = imdilate(lines,se);

%show line alone

imshow(lines)

%show lines overlaid orginal image

imshowpair(lines,img)

Hope it helps.