1351954 - Right ventricle segmentation using a narrowband level set method in cardiac MRI

Compared to the segmentation of the LV in cardiac MRI, segmenting the RV cavity is a more challenging task due to several reasons [1, 2]. First, the in-flow blood and partial volume effects blur the RV cavity borders. Second, the shapes of the RV usually vary significantly for different slices and different pathologies. Third, the RV trabeculation is much more prominent compared to the LV. Due to these challenges, RV segmentation often needs extensive manual correction of automated methods or a totally manual approach in some pathologies, which is time-consuming. The level set method [3] is an effective approach for image segmentation due to its ability to catch complex topological changes effectively. We propose a new RV segmentation model using the narrowband level set method in this work.

Methods: In this work, we propose to adapt the level set method for RV segmentation. We assume that the segmentation curve c is given by the zero level set of a high dimensional function φ: c = {(x,y)|φ(x,y)=0}.

This method can effectively catch the topological changes in the shape to be segmented. With the evolution of the level set function starting from some initial level set function, the segmentation curve will move implicitly and finally fit the shape for segmentation. To reduce the computational complexity of the calssical level set method, we propose to perform the evolution of the level set function in the neighborhood of the RV only in this work. We term this method as narrowband level set method.

To adapt the narrowband level set method for RV segmentation in cardiac MRI, we assume that the level set function u is a general piecewise smooth function. We can then show that we can approximate the level set functional with a functional whose gradient flow equation can be modeled as a modified Allen-Cahn [4] equation: ∂u/∂t=M(εΔu-F'(u)), where M and u are two user-defined parameters and F = 0.25(1-u2)2 is a double-well potential function.

During the segmentation, the user is directed to draw any initial curve around the RV as the initialization. Based on the initial curve, it automatically created a narrowband around the initial curve in the image domain for the level set function evolution.

Results:

We first show the results from the simulation study. The results from the simulation study are shown in Fig. 1.

We then use the proposed narrowband level set method to segment the RV in the short axis slices from two 16 years old male subjects: one normal subject and one with Tetralogy of Fallot. The images were acquired using the bSSFP sequence. In Fig. 2 and Fig. 3, we showed the segmentation results from the diastole phase and systole phase. 

Conclusion:

In this study, we proposed a narrowband level set method to segment the RV in cardiac MRI based on a modified Allen-Cahn equation. Experimental results indicate that the proposed RV segmentation scheme is able to provide good segmentation results in a short time.