The human body surface has a property of right-left symmetry. Some anatomical sites, such as breasts, legs, and arms, also possess rotational symmetry. This geometric symmetry constitutes a difficult technical barrier for 3D scanner-based optical surface imaging systems. Such systems only measure the distance between the interrogation light source and points of interest on the body surface. Because of this, these systems are unable to differentiate those points on the body surface that move along the symmetry direction or rotate about the symmetry axis simultaneously. This technical limitation hinders their applications on symmetric body surfaces. To overcome this technical barrier, we have developed an algorithm which can calculate a surface-specific region of interest (ROI) for accurate surface tracking. Our algorithm consists of 6 steps: 1) remove background noise; 2) create a seed ROI; 3) add a maximal margin to the seed ROI; 4) rectify ROI overexpansion; 5) eliminate flat topography; and 6) maximize asymmetry and height differences of ROI. Briefly, the algorithm starts by cutting off irrelevant surfaces, including floor, couch, head, lower body, and immobilization devices. A square seed ROI is computed and centered on isocenter, assigned an initial size of 80x80 mm in x and y-axis. The algorithm searches for the maximal and minimal height (z-axis) within it. Thereafter, a 200 mm growth margin is added to the seed ROI. The algorithm compares z-values of two neighboring points in the new ROI. If they are significantly different, then there is a change in topographic feature. ROI grows and repeats the same operation along this direction. The algorithm stops if z-values are too close. The algorithm may expand into body lateral side. To avoid it, a z-value thresholding technique is used so that points with a height lower than the set threshold are eliminated. Two additional thresholding techniques are employed to eliminate a surface with very flat topography. Finally, a symmetry break-up loop is used to maximize the asymmetry and height standard deviation of the ROI. To speed up algorithm performance and circumvent unexpected noise features (such as mask and sheet), the idea of contour line is used in the algorithm. With the ROIs generated by our algorithm, it was found that our optimized asymmetric ROIs were more effective and accurate in detecting shifts by symmetric body surfaces than using symmetric ROIs. It is particularly effective and sensitive in detecting rotational shifts by body surfaces with a rotational symmetry. For optical surface imaging systems, poorly defined large ROIs may produce stable, but often wrong tracking results. This problem is particularly pronounced for symmetric surfaces, such as the breast surface. Our novel concept of optimized asymmetric ROI with a limited size is extremely valuable for real-time patient local tracking applications.
Read full abstract