Accurate segmentation of the left ventricular wall in echocardiography of murine models is crucial for cardiac functional assessment, but it presents a challenge due to anatomical variability, low resolution and the effort required in manual annotation. This work proposes a modular automatic segmentation architecture composed of two modules: the Res-SE-U-Net, which is based on U-Net architecture and incorporates a ResNet18 encoder and Squeeze-and-Excitation attentional blocks to extract spatial features, followed by a ConvLSTM module to refine the segmentations by incorporating temporal information. The model was trained and evaluated on a dataset of 452 long-axis echocardiography frames, with data augmentation applied during training, achieving competitive performance metrics, with an Intersection over Union score of 0.803 and a Dice coefficient of 0.889, which proved to be an efficient, reproducible and effective approach.