In echocardiography, accurate automatic left ventricle wall (LVW) detection is a prerequisite for the diagnosis of most heart diseases. However, there exist crucial challenges in B-mode echocardiography segmentation due to the myocardial anisotropy and the backscattering effect. In this paper, we propose a neural network-based LVW segmentation method inspired by the camouflaged object segmentation (COS) technique. The selected difference (SD) scheme is proposed, which enhances the accuracy and robustness of LVW segmentation by utilizing complementary features of radio frequency data and B-mode image data jointly. Evaluations are performed using 200 test phantoms in the Dice coefficient metric. The proposed scheme outperforms the conventional LVW segmentation baseline neural network that employs b-mode images only.