My White Matter Tina Kapur

Tina Kapur
tkapur"at"alum.mit.edu

Goal: Change Detection in Breast Tissue

Our goal is to develop techniques for segmentation and registration of deformable 3d structures such as the breast tissue. Successful automatic registration of a time series of breast images of a patient can be used for early detection of changes in the breast tissue that could be highlighted for closer inspection by the radiologist. Our conjecture is that the goal of registration will be facilitated by the segmentation of anatomical structures that stay stable over time, and hence can serve as "landmarks" or "fixed points" in the registration process. Alternatively, a coupling of the registration and segmentation processes could also prove effective.

Motivation

1 in 8 women in the US develop breask cancer and 1 in 33 of those die every year.




Some Results

We register pre- and post-gadolinium MRI of the same patient using mutual information of feature points. The class of transformations allowed is global affine. Work on local affine tranformations is under progress.

The resolution of this data is 256x512x30 pixels (second dimension is actually 364, but gets zero padded) or 300 microns x 600 microns x 2 mm.

Initial Unregistered Images

The three rectangles on the left half of this image show the three cross-sections of the pre-gad MRI. The right three rectangles show the corresponding cross-sections of the post-gad MRI of the same patient, taken on the same day.

Final Registered Images

The three rectangles on the left half of this image show the three cross-sections of the pre-gad MRI (same as in the previous image). The right three rectangles show the corresponding cross-sections of the post-gad MRI after our feature based, affine, MI registration has been performed on it. Notice that the tumors that were not so bright in the pre-gad image show up much brighter in the post-gad image.

Here is an MPEG of this registration.

Here is an SGI movie of this registration.

Here is an MPEG of another registration.

Intensity Distribution of Registered Images

The top image shows the intensity distribution of one slice of the the pre-gad scan, and the bottom image shows the intensity distribution of the corresponding registered post-gad slice. The brightness peaks signify substantial absorption of the contrast agent, and hence indicate the presence of tumor. A simple image differencing can isolate such areas.

Difference Image to Highlight Enhanced Areas

Tina Kapur's Home Page