Shrinkwrap - DeformableSurface program

We have developed and implemented a deformable surface model (Computer Vision and Graphics in Fluorescence Microscopy, LM Lifshitz, KE Fogarty, JM Gauch, E Moore, SPIE vol. 1808, Visualization in Biomedical Computing, pp. 521-534, 1992) which has been used numerous times over the years to locate the plasma membrane of fluorescently labeled smooth muscle cells. This model is composed of vertices spaced along a model surface which is initially placed within a 3D image, near its correct location. Each vertex is then allowed move in response to image intensity gradients, attempting to move to locations in the image with a higher intensity (which are typically labeled channels, but possibly other structures such as locations with strong edge strength). The position of each vertex also imposes a "cost" based upon how much curvature its position would impose onto the surface; Vertices are moved to maximize a function of the image intensity at their location minus a measure of this curvature. Our implementation calculates curvature measures (i.e., first and second derivatives of the surface) and intensity measures locally and quickly. This results in a fast algorithm which is also easy to modify to handle additional local constraints. For instance, it is simple to freeze certain vertices in space so they cannot move (e.g., if their initial position is known to be very accurate) or to restrict motion of the vertices to a plane (e.g., if a 2D time series is analyzed as one "3D" image).

Once a surface (e.g., a plasma membrane Deforming Surface ) has been defined we have numerous software modules to facilitate analysis of fluorescence relative to that surface. One module eliminates all signal in an image which is not within a defined distance from the surface. This has been used to remove non-specific label inside a cell so that analysis can be restricted to label which is on, or very near, the plasma membrane. Another module can map an interior point to the nearest location on the surface. This enables, for example, finding the distance between a molecule inside the nucleus and the nuclear membrane (e.g., if it might be exiting via nuclear pores). Another module creates a histogram of fluorescent intensity inside a cell as a function of distance from the plasma membrane (e.g., to enable the application of a threshold which varies within the cell based upon distance from the plasma membrane).