Breast Tumor Image Segmentation Based On Fuzzy Logic Attention Mechanism U-net

As a malignant disease that affects women’s health,breast cancer has always been diagnosed by medical self-examination and doctor’s experience with medical imaging,which has caused a lot of burden on patients and doctors.With the development of artificial intelligence machine learning and the popularity of deep learning of neural networks,more and more computer methods have been applied to medical clinics.Image segmentation in this field is an important research content in the field of image processing,and has broad application prospects in medical image processing.Starting from this direction,the paper studies and proposes a new deep learning algorithm model for segmentation of breast tumor ultrasound images based on fuzzy logic attention mechanism.Although the traditional image segmentation algorithm achieves high efficiency of simple execution,it is difficult to achieve ideal results for the image segmentation of the medical image,which is characterized by complex texture edges.For this phenomenon,this paper mainly does the following work:(1)For the original breast tumor ultrasound image,the histogram equalization and wavelet transform method are used to make the detail features clearer and have more information.The U-Net model with significant medical image segmentation effect is improved by the attention mechanism,and the upper part is sampled separately.And the attention module embedded in the feature channel and the spatial region in the downsampling structure,the former pays more attention to the importance degree between the feature channels in the same size structure,and the latter pays more attention to the importance weight of different elements in each position region on the same feature map.This way,the extracted information can be further utilized,and the shallow and deep information fusion of the context makes the segmentation result more accurate.(2)Innovatively introduce fuzzy logic into the neural network,combine with the method of recalibrating the importance degree of the feature elements in the attention mechanism,and calculate the uncertainty of each weight through the membership function to make it measure from the determination.The feature importance weight is transformed into the weight with fuzzy logic uncertainty.At the same time,the replacement loss function is introduced,and the strategy of cavity convolution and residual structure deepening network is introduced to further improve the segmentation effect of the model on breast tumor.