Early Diagnosis And Screening Of Breast Cancer Based On Multimodal Image

Ultrasound(US)and mammography are the most common and routine techniques for early screening of breast cancer.In ultrasound examination,B-mode ultrasound and Shear-wave Elastography(SWE)complement each other in the diagnosis of breast cancer,which can help diagnose the benign and malignant of breast tumors.Breast density grading changes during breast cancer development is an important signal,and correct grading can predict breast cancer development.This article uses deep radiomics methods,combined with multimodal data(including Bmode ultrasound,Shear-wave Elastography,and mammography)for early diagnosis and screening of breast cancer.The B-mode ultrasound image can be acquired while patients doing shear-wave elastography examination.Firstly,breast tumors were detected in B-mode ultrasound images by means of object detection algorithms.According to the correspondence between B-mode ultrasound and shear-wave elastography image,we obtained the region of interest of breast tumors and surrounding based on shear-wave elastography images.Different from tumor segmentation methods,our method includes breast tumors and surrounding tissues.The surrounding tissues provide more abundant information and features for the diagnosis of benign and malignant breast tumors.The ROIs of breast tumors based on shear-wave elastography ultrasound images were input to deep convolutional neural networks(DCNN)to extract highthroughput features and classifiers for benign and malignant classification.Our algorithm has an accuracy of 95.8%,with a sensitivity of 96.2% and a specificity of 95.7% for breast tumors classification.Compared with other algorithms,the algorithm has effectively improved the diagnostic accuracy.Breast density grading based on mammography is an important indicator in early screening of breast cancer.Different grades of breast density directly affect the risk of breast cancer.Therefore,it is important to correctly determine breast density grading for early screening of breast cancer.Deep convolutional neural networks extract high-throughput,high-level,and high-abstract features from images for breast density grading.Finally,the accuracy of four grades classification task reached 92.6%.It is worth noting that for the simplified two grades classification task,our method has achieved an accuracy of 100.0% for one of the grades.In summary,the main content of this paper can be divided into three parts: the first part is based on B-mode ultrasound image for breast tumors detection;the second part is based on elastic ultrasound image of breast tumors benign and malignant diagnosis;the third part is based on mammography for breast density grading.These three components are combined for early diagnosis and screening of breast cancer.