Author : Mr. Shao Fan
Last Updated : 1 October 2004

3D Prostate Surface Detection in Ultrasound Images

Introduction

Prostate boundary detection in ultra-sound images plays a key role in prostate disease diagnoses and treatments. However, due to the low contrast, speckle noise and shadowing in ultra-sound images, this still remains as a difficult task. Currently, prostate boundary detection are performed manually, which is arduous and heavily user dependent. A possible solution is to improve the efficiency by automating the boundary detection process with minimal manual involvement. This poster presents a new practical method based on level sets to semi-automatically detect the prostate surface in 3D transrectal ultrasound (TRUS) images.

Methods

In level set formulism, the volume segmentation problem can be expressed as the computation of a 3D surface S(t) propagation in time along its normal direction, and the propagating front S(t) is embedded as the zero level set of a time-varying 4D function :

The level set method then evolves the 4D function that contains the embedded motion of S(t) as

where speed F can be a function of the front characteristics (e.g. the curvature), and the image characteristics (e.g. gradient). Figure 1 shows a diagram of 2D curve propagation with the level set method.

Figure 1. Example of 2D curve propagation with the level set method.

However, due to the poor quality of ultrasound images, the boundary feature of the object is usually not salient enough and the image gradient information is weak. It usually causes the “boundary leaking” problem as shown in Fig.2 when we apply the level set method to detect the 3D prostate surface.

Figure 2. An example of “boundary leaking” problem of level set method. White curve: the detected boundary of the level set method; Green curve: the manually outlined boundary.

To remedy this problem,

	first, we roughly extract the prostate region R by a fast discrimination method according to the intensity likeness;
	then we use Gaussian mixture model (GMM) to describe the intensity distribution inside the prostate so that we can design an indicator function P to determine whether or not the intensity mainly contributes to the prostate;
	the region information and Gaussian mixture model, instead of image gradient, are then integrated into the level set method to design the new speed function F_new. Meanwhile, the gradient flow is also added to the level set method to further improve the algorithm. That is,

Results

We applied the proposed method to eight typical 3D TRUS images to detect the prostate surfaces. The initial surfaces (level sets) are manually generated in particular slices on the condition that the level sets should be placed totally inside the prostate. Fig.3 illustrates an example of this multiple level-sets initialization and Fig.4 gives the detected result.

Figure 3. An example of multiple level-sets initiali-zation. (a) 5 evenly separated slices, (b) 2 level sets at slice 42, (c) 3 level sets at slice 84, (d) 3 level sets at slice 126, (e) 1 level set at slice 168.

Figure 4. Result obtained from the user-defined initial level sets shown in Fig.3. (a) Original image (252×194×256), (b) detected surface, (c) 3 sets of contours in 2D cross-sectional images.

Conclusion

A multiple level-sets model for prostate surface detection in 3D transrectal ultra-sound images has been developed. Region information, statistical distribu-tion model as well as the gradient flow are well integrated into the level set method to improve the algorithm’s performance. The results have shown the effectiveness of our proposed method.

Publications related to 3D Level Set.

We would be glad if you could sign our guest book.

For more information, please contact the principal investigator: