Research On Microwave Imaging Of Breast Tumors Based On Machine Learning

In recent years,the incidence of female breast cancer in China has increased year by year,and it has become a major problem that threatens women’s health.Early detection can improve its cure rate.Statistics show that the cure rate of breast cancer in China is significantly lower than in developed countries.Breast tumor screening can detect early breast cancer,so that it is a focus issue of medical imaging.Existing studies have shown that normal breast tissue and malignant tumor tissue have different electromagnetic response characteristics,which provides a theoretical basis for the application of microwave imaging in breast tumor screening.Microwave imaging has the advantages of no radiation and high imaging resolution,which makes it widely used in breast tumor screening.This paper focuses on two issues.Firstly,a microwave imaging algorithm based on genetic algorithms is proposed to image breast tumors with regular shapes.Secondly,improve the imaging algorithm to enhance its ability to image actual breast tumors with complex structures.This paper uses microwave tomography to realize the imaging of two-dimensional tumors,and uses the finite-difference time-domain method to solve the problem of microwave imaging forward scattering.The inverse scattering problem of microwave imaging is a nonlinear and ill-conditioned problem.This paper uses genetic algorithm to solve the inverse scattering problem of microwave imaging.This paper proposes a coding method for contour extraction.This approach can achieve high-resolution imaging while shortening the length of the chromosome of the genetic algorithm,thereby reducing the amount of calculation of the genetic algorithm.At the same time,adding distance compensation to the fitness function reduces false optimal situations.And this paper constructs an iterative way of dynamic contour extraction,which improves the iterative efficiency of genetic algorithm.Simulation experiments have confirmed that the imaging algorithm proposed in this paper can achieve high-resolution imaging with a resolution of 1mm for two-dimensional breast tumors.In the meantime,it is also confirmed that the coding method of contour extraction and the iterative method of dynamic contour extraction can improve the imaging accuracy and the iterative speed of genetic algorithm.In this paper,the K-means clustering algorithm is used to transform the actual breast tumor ultrasound map into a binary grid map.Based on this binary grid map,an actual twodimensional tumor model with complex contours is established.Next,this paper improves the imaging algorithm to enhance the imaging ability of actual tumor models.In the first place,the logarithmic transformation is performed on the distance compensation fitness function to improve the iterative speed of the genetic algorithm.In the second place,ensemble learning is applied to microwave imaging of breast tumors,which improves the ability of genetic algorithm to optimize microwave imaging of irregular breast tumors.Ensemble learning adopts Bagging’s ensemble idea to construct a base learner,and the combination method is a soft voting method.In this paper,simulation experiments prove that the improved imaging algorithm has a good generalization ability,which can realize imaging of breast tumors with different contours,different positions and different sizes.The imaging algorithm in this paper can achieve 1mm resolution ratio for breast tumors with a diameter of about 1cm,and the precision ratio and the recall ratio of the imaging results are both above 96.97%.