ChapterZero

I’ve been working on a digital contrast agent for detecting early signs of early signs of arthereosclerosis non-invasively through 3d CT scans(currently impossible), with Dr. Papadakis (probably lately, the most mentioned name here ), since September of this year. The technique we’re using is a combination of stochastic modeling and wavelet decomposition. Specifically, we apply a nonseparable wavelet transform to the entire 3d volume scan of a coronary artery, and assume that each individual tissue corresponds to a texture, which we consider to be random field. We assume that the individuality of each texture is preserved across the scales. We require the end user of the algorithm to provide us with a small region of the volume that is known or assumed to be ‘normal’ tissue, and fit a linear predictive filter over each subband of this image. By running these filters over the entire volume, and considering the difference between the actual and predicted wavelet coefficients, we can isolate the ‘abnormal’ portions of the image, and reconstruct these.

Today we had a major breakthrough: we have what looks like the first evidence of the algorithm differentiating between classes of soft tissues. Until now, we were having much more success at locating calcium that anything else, but that is of limited use, since current thresholding techniques can already isolate calcium deposits. I am very impressed with these results! See the magic for yourself (below the fold: the images are huge).

The top image is the original CT data, which has no visible difference between types of tissues, the second is the actual histology, where you can observe several large lipid pools, and the bottom is the CT image processed using our algorithm.

Possibly relevant posts:

• A Possible Future (6/3/2005)
• But that’s just least squares… (8/9/2006)
• Inpainting (11/18/2005)