University of Houston

Breast reconstruction is an important surgical component for many women undergoing breast cancer treatment. The purpose of breast reconstruction is to recreate a breast form that is satisfying to the patient, facilitating her psychosocial adjustment to living as a breast cancer survivor. Breast reconstruction is usually completed in a multi-step process which can take up to a year or longer. The goal of this project is to develop algorithms to monitor and quantify changes in local breast morphology during the reconstruction process.

This project will encompass the following three topics:

1. Align 3D torso scan of individual patients from multiple-visits to achieve correspondence between the images. In addition to breast surgery related anatomical changes, the multi-visit images from the same patient acquired at different clinical visits may also change as a result of variations in the (a) object coordinate systems due to differences in patient positioning and posture; and (b) patient's BMI due to physiological weight changes. As a first step, registration of the multi-visit images is required in order to analyze images in the same coordinate system, and to facilitate quantification of local morphological changes occurring over time in the operated breasts.
2. Automatically extract the breast mound surface regions from images. In this step, an automated algorithm is needed to extract the breast data in order to overcome the operator bias inherent to manual segmentation.
3. Quantitatively analyze the local morphological changes of breasts. For each point of the breasts, the displacements and deformations of the surface features along the X, Y, and Z directions between different visits will be calculated.

For the first step, we have developed a semi-automated rigid registration algorithm. The approach relies on the assumption that while the soft tissues of the patient's body may change over time, the skeleton is relatively stable. Thus the transformation of skeletal frame can be treated as being rigid. Selecting points with reference to the skeletal frame and maximizing the correspondence between these points can then achieve 3D image registration. This algorithm is robust to discrepancies in fiducial point selection (sternal notch and umbilicus) of about 2 cm.

In future, we propose to develop an automated registration algorithm of 3D torso images in order to avoid operator bias in the semi-automated method. Finally, we will design and develop algorithms for breast data extraction and quantitative analysis of local breast morphology changes.