G-UNet: A Breast Cancer Lesion Segmentation Method Based on Cross-Attention

To address the issues of varying sizes and shapes, as well as blurred boundaries of breast cancer lesions in magnetic resonance imaging, G-UNet segmentation algorithm based on a cross-attention mechanism is proposed to enhance segmentation accuracy and reduce misdiagnosis. This algorithm leverages the efficient feature representation of GhostNet and the multi-scale feature fusion capability of UNet, while employing a cross-attention mechanism to intensify focus on the breast cancer lesion areas, thereby more precisely capturing the details and boundary characteristics of the lesions. This approach effectively tackles the complexity of breast cancer lesions in MRI images, leading to a notable improvement in the precision and stability of segmentation results. Compared to UNet, G-UNet achieves an increase of 4.19 percentage in intersection over union and 0.21 percentage in Dice coefficient, respectively. Experimental results demonstrate that the proposed algorithm can elevate the accuracy of cancer lesion segmentation and effectively reduce the misdiagnosis rate in image-based diagnosis.