An SPCNN-GVF-based approach for the automatic segmentation of left ventricle in cardiac cine MR images

Abstract

Purpose

Accurate segmentation of left ventricle (LV) is essential for the cardiac function analysis. However, it is labor intensive and time consuming for radiologists to delineate LV boundary manually. In this paper, we present a novel self-correcting framework for the fully automatic LV segmentation.

Methods

Firstly, a time-domain method is designed to extract a rectangular region of interest around the heart. Then, the simplified pulse-coupled neural network (SPCNN) is employed to locate the LV cavity. Different from the existing approaches, SPCNN can realize the self-correcting segmentation due to its parameter controllability. Subsequently, the post-processing based on the maximum gradient searching is proposed to obtain the accurate endocardium. Finally, a new external force based on the shape similarity is defined and integrated into the gradient vector flow (GVF) snake with the balloon force to segment the epicardium.

Results

We obtain encouraging segmentation results tested on the database provided by MICCAI 2009. The average percentage of good contours is 92.26 %, the average perpendicular distance is 2.38 mm, and the overlapping dice metric is 0.89. Besides, the experiment results show good correlations between the automatic segmentation and the manual delineation (for the LV ejection fraction and the LV myocardial mass, the correlation coefficients R are 0.9683 and 0.9278, respectively).

Conclusion

We propose an effective and fast method combing the SPCNN and the improved GVF for the automatic segmentation of LV.

Appendix

As shown in Fig. 9, the single neuron structure of SPCNN model includes three parts: input field, modulation field and pulse generator field [28]. And, the discrete SPCNN was described as follows:

$$\begin{aligned} F_{xy}[n]= & {} S_{xy} \end{aligned}$$

(17)

$$\begin{aligned} L_{xy}[n]= & {} {\sum _{({kl}){\in }N_{xy}}} W_{xykl} Y_{kl} [n-1] \end{aligned}$$

(18)

$$\begin{aligned} U_{xy}[n]= & {} F_{xy}[n](1+\beta L_{xy}[n]) \end{aligned}$$

(19)

$$\begin{aligned} Y_{xy} [n]= & {} \left\{ {\begin{array}{ll} 1&{}\hbox {if}\quad U_{xy}[n]>E_{xy}[n-1] \\ 0&{}\hbox {otherwise} \\ \end{array}} \right. \end{aligned}$$

(20)

$$\begin{aligned} E_{xy} [n]= & {} e^{-\alpha _e}E_{xy} [n-1]+V_E Y_{xy} [n-1] \end{aligned}$$

(21)

In the input field, each neuron $N_{xy}$ in the position (x, y) receives information from both the feeding input $F_{xy}[n]$ and the linking input $L_{xy}[n]$ in iteration n. $F_{xy}[n]$ is equal to the normalized gray intensity of the original input image $S_{xy}$. $L_{xy}[n]$ is the result of that $N_{xy}$ communicates with its eight neighboring neurons $N_{kl}$ through a constant synaptic weights W. $Y_{kl}[n-1]$ is the output of $N_{kl}$ in the previous iteration. In the modulation field, $F_{xy}[n]$ and $L_{xy}[n]$ are modulated through the linking strength $\beta$ to obtain the internal activity $U_{xy}[n]$ which is used to judge whether the neuron $N_{xy}$ fires or not. In the last (pulse generator) field, $U_{xy}[n]$ is compared with the dynamic threshold of previous iteration $E_{xy}[n-1]$. If $U_{xy}[n]>E_{xy}[n-1]$, $N_{xy}$ fires (i.e., setting the output $Y_{xy}[n]$ to one); otherwise $N_{xy}$ does not fire (i.e., setting $Y_{xy}[n]$ to zero). In this way, the output matrix Y[n], which consists of 0 and 1, generates a binary image in iteration n. The dynamic threshold $E_{xy}[n]$ is updated in each iteration according to the output $Y_{xy}[n-1]$: If $N_{xy}$ fires in the previous iteration $(Y_{xy}[n-1]=1),E_{xy}[n]$ will increase by parameter $V_{E;}$ if $N_{xy}$ does not fire ($Y_{xy}[n-1]=0$), $\hbox {E}_{xy}[n]$ will decay by an exponential decay coefficient $e^{-{\alpha }_e}$ [25].

Fig. 9

Single neuron structure of SPCNN model [28]