Computer Vision System
The goal of our Computer Vision system is to automatically recognize and outline anatomical or pathological structures in medical images so that they can be analyzed quantitatively.
We have developed a generic architecture that achieves this by automatically segmenting a given medical image and matching extracted image primitives to anatomic structures in a model.
Methodology
The system uses a unique approach and computer architecture for model-based segmentation of medical images, currently focussing on computed tomography (CT) images [Brow97a, Bosc02a]. The key components of the architecture are an anatomical model, an inference engine and image processing routines. Segmentation involves matching objects extracted from the image to anatomical objects described in the model. Image and model objects are matched by transforming them into a common parametric feature space for comparison. Knowledge of the expected size, shape, topology and X-ray attenuation of anatomical structures are stored as features in a model. These features are used to guide 3-D segmentation of the modeled anatomy. The image segmentation engine extracts regions of interest (ROIs) from the image using seeded region-growing and morphology. The ROIs are matched to the anatomical structures in the model by comparing features using fuzzy logic. The model-based segmentation approach is both powerful and flexible. By using different anatomical models it has been applied to a variety of problems [Brow98a, Brow99a, Brow00a, Brow00b, Brow01a, Brow01c].
Applications of the computer vision system to: (a) lungs and main stem bronchi; (b) pulmonary vessels; (c) kidneys; and (d) neurovasculature