Segmentation of Mammogram Images Using U-Net with Fusion of Channel and Spatial Attention Modules (U-Net CASAM)

One of the most critical concerns that must be addressed worldwide is early breast cancer diagnosis because it can help the patients with a higher survival rate. Early detection of breast cancer can significantly lower treatment expenses; mammograms have been used to identify the disease. To plan treatments, segmentation assists doctors in measuring the amount of tissue in the breast. This work uses a U-Net model combining spatial attention and fused channel modules. The U-Net CASAM module is trained with the CBIS-DDSM dataset. The proposed segmentation algorithm’s performance is assessed using the combined INbreast and MIAS datasets regarding intersection over union (IoU) and Jaccard index.