3-D/4-D Segmentation of Structures in CT Images for Radiotherapy Planning
Contouring anatomic structures at risk in thoracic CT images is critical for radiotherapy treatment planing. Since manual contouring is labor intensive, considerable attention has been devoted to automating the process.
3-D Esophagus Segmentation in CT Images:
Locating the esophageal wall is particularly challenging. Inherent difficulties in segmentation of the esophagus include the lack of consistent intensity contrast and variable appearance in regions which contain air bubbles or remains of oral contrast agent. To locate the esophagus in thoracic CT scans we use a variational framework. To address challenges due to low contrast, several priors are learned from a training set of segmented images. Our algorithm first estimates the centerline based on a spatial model learned at a few manually marked anatomical reference points. Then an implicit shape model is learned by subtracting the centerline and applying PCA to these shapes. To allow local variations in the shapes, we propose to use nonlinear smooth local deformations. Finally, the esophageal wall is located within a 3D level set framework by optimizing a cost function including terms for appearance, the shape model, smoothness constraints and an air/contrast model. (ICPR’10)
Anatomy Based Spatial Model:
Using the esophagus centerline location with respect to some neighboring structures that are easy to segment, we built a spatial model of esophagus centerline from annotated training data sets. The probability distributions of dx(z), the normalized distance of centerline x (anterior-posterior) coordinate with respect to 2 neighboring structures were computed. dy(z) was computed similarly.(ISBI’10) (ICCR’10)
Locally Smooth Deformations to Improve Global Linear PCA based Shape Model:
The global shape model was calculated by applying PCA to annotated esophagi after centerline subtraction, i.e. shifting center of each slice to the origin.The shape model is a linear combination of k most important PCA modes added to the mean shape.
Before applying the shape model, the test data needs to be centered around origin. Therefore, first a centerline estimation is performed.
Due to high data variability in x-y direction and rough center estimation, we include into our shape model a local transformation function that acts on x and y directions. The local deformations allow the algorithm to correct for the inaccuracies in center estimation (b) and better capture local variabilities that cannot be represented by the global PCA-based shape model. We choose N uniformly sampled action points zk through the centerline. We allow a local transformation, in the form of a translation in x-y plane, to be applied to each zk. This translation affects the neighboring slices and this effect smoothly dies off as one moves away from the action points in z-direction. Such a local deformation A can be formally defined as follows:
where ak, bk are the amount of deformation at action points in x and y directions respectively. (ICPR’10)
Initial centerline estimation: x
Final center location after locally smooth deformation: +
