HUSE: Squeeze and Excitation based Half-Unet for left ventricle MRI segmentation

In this study, presents a Half-Unet, that integrates Squeeze and Excitation (SE) attention mechanism within the Ghost Module (GM) to enhance medical and non-medical image segmentation. The proposed architecture enhance the strengths of the basic Unet while integrating SE module to recalibrate channel-wise feature responses, thereby improving feature representation. Model was assessed using the MICCAI 2009, which consisted of magnetic resonance imaging images, re-sized to 256 x 256 pixels. Computable results demonstrate that the Half-Unet Attention model surpasses both the basic Unet and Half-Unet models, by achieving a precision of 0.7714, recall of 0.6764, specificity of 0.9968, IoU of 0.5634, and an overall score of evaluation metric is 0.7321. Visual comparisons further confirm the superior segmentation accuracy and detail captured by the proposed model.