Idea Highlight B: Shape-Based Segmentation

Research at the Engineering Research Center for Computer-Integrated Surgical Systems and Technology is enhancing the ability of clinicians to plan and execute minimally invasive interventions, by greatly improving the important first step of “segmentation”: the process of automatically detecting and accurately outlining the image of the organ that is the object of the procedure.

Robust and reliable automatic segmentation has been very challenging to date, because of insufficient information and undesirable “noise” in the images being used. When the image contrast in the segmentation is weak, clinicians rely on their knowledge of anatomy to make up for these deficiencies. Better use of computational modeling of anatomical shapes, and their variability in the population, can significantly improve the quality of the automatic segmentation.

The research team at MIT has constructed the statistical model of shape from previously segmented scans, by estimating the mean shape and the principal modes of its deformation within the population from which the image data was collected. The left-side images in the figure illustrate such a model for the prostate (red), the colon (green) and the surrounding muscles (yellow). The array of 3D surfaces illustrates the mean shape (center column) and its variation (left to right) along the four most “important” directions estimated from the data. In this example, the model captures not only the organs’ shape, but also their relative position and orientation. It is then used during the segmentation of new scans to constrain the shape of the hypothesized segmentation. The intensity information in the adjacent structures leads to a robust segmentation of the prostate, which would be very challenging if the prostate were considered in isolation.

The right-side images in the figure show a novel scan that was automatically segmented using this approach, as well as the 3D rendering of the segmented structures.

This automatic procedure greatly facilitates planning and navigation for minimally invasive treatment of the prostate cancer. Moreover, the applications of the segmentation algorithms developed and demonstrated in this program range from surgical planning and navigation to population-based studies of neuroanatomy and the effects of diseases on the morphology of the brain.


<< back to Highlights main menu