Classification Of Benign And Malignant Tumors Based On Deep Learning Using ABUS Sequence Images

Due to ABUS(Automated Breast Ultrasound)imaging with the way of automatic scanning towards breast,a doctor or researcher can get a series of ultrasonic pathological image sequences with the same size.Therefore,compared with hand-held ultrasound images with different sizes,it has unique advantages for the application of artificial intelligence technology recognizing ABUS images to screen benign and malignant breast tumors.In order to assist imaging doctors in determining the benignity and malignity of breast tumors more accurately and accelerating the speed of analyzing breast ultrasound images,this paper proposed originally a method based on deep learning to classify benign and malignant breast tumors in ABUS sequence images.At first,we will design the Preprocessing Model to eliminate the redundant information in the ABUS sequence images.The Preprocessing Model consists of the Extract-IOI model and the Cropping-ROI model.The IOI(image of interest)sequence images are extracted by Extracting-IOI model from the ABUS sequence images.The Cropping-ROI model crops out the ROI(Region of Interest)from the IOI sequence images.Then,we design the original SDCB-Net(Shallow Dilated Convolution Branch Network).SDCB-Net combines with VGG16 pretrained model to build SEF-Net(Shared Extracting Feature Network)SEF-Net can extract feature of ROI sequence image.SEF-Net which includes SDCB-Net can avoid the overfitting problem caused by the difficulty in obtaining enough medical images with labels.On this basis,we designed GRUC-Net(GRU Classified Network)to extract and integrate the correlation features among ROI sequence image features,and finally realizing the benign and malignant classification of tumors in ABUS sequence images.The experimental results show that the accuracy and AUC value of ABUS sequence images on the training set are 93.60%and 99.43%,respectively.In the test set,the results reached 92.86%and 97.21%,respectively.It can be seen that the application of this method to distinguish benign and malignant tumors of ABUS sequence has high performance,improving the speed and efficiency screening of breast tumor to a large extent.Therefore,it can help more patients timely diagnose breast tumors and enhance the clinical application significance of ABUS.