NOTES.txt

Notes:
	- 100 data sets!
	- 100 - 450 slices apiece
	- start right below throat, end right below lungs
	- voxel size is a constant for each data set
	- ranges from 0.5 to 7.0 mm (per side; x = y = z)
	- input: 512x512 {bmp, pgm}
	- output: 512x512 pgm

(Empirical) Classifications (vox = voxel):
	- B(kgd)-vox = [0, 50]
	- C(hest)-vox = [150, 255]
	- L(ung)-vox = [25, 150]
	- N(oise)-vox = 255


--------------------------------------------------------------------------------
List of steps | detail

1. (possible) 3d median smoothing
	-- for every voxel, check 27 neighbors and sort them.
	-- pick the median and replace the initial voxel with its value
	-- this makes the quality of the region-growing better
	-- but at the expense of blurring or removing significant data

2. Get rid of B-regions in the lung
  -- run 3d region growing with seeds on the 4 edges of the root slice
  -- it must mark all B-regions as 255
  -- there is a fine line between B and C regions, but region growing should be able to crawl around it by trying to find a breach in the line in any slice

3. Get rid of the chest region (or as much of it as possible)

4. isolate the lung region as a black region in a new volume;
	region grow the entire volume;
		store white pixels -> new volume
		skip black ones (the new volume is initialized to black anyways)
		for every white pixel in the new volume, run the mgrf

5. the new volume contains the boundary for the lung