Research On The Method Of Breast Imaging Lesions Localization And Diagnosis Of Benign And Malignant Based On Neural Network

The incidence and mortality rate of breast cancer is increasing and it is the leading disease threatening of women health.So,early detection and treatment can effectively reduce the mortality rate of breast cancer patients.Breast mass and calcification are important manifestations of breast cancer.However,as the low pixel resolution and low contrast between breast tissue and lesions in medical images,doctors may have a high misdiagnosis rate for treatment and diagnosis.In this paper,we focus on breast mass segmentation and determination of benign and malignant breast lesions.Additionally,we review related literature,analyze and discuss a combination method of Pulse Coupled Neural Network（PCNN）and Convolutional Neural Network（CNN）to improve the accuracy of breast mass segmentation and determination of benign and malignant lesions.Image segmentation and image classification are important research directions in medical image processing.Automatic segmentation and classification of medical images combined with artificial intelligence play an important role in the early diagnosis of breast cancer disease,and the segmentation and classification accuracy of medical images are extremely important to assist clinicians in accurate diagnosis and treatment.Our works are shown in subsequent steps:（1）Due to the complex information of mammogram images,mass segmentation is easily affected by high-density glands and tissues.For the mass characteristics in mammogram images,by analyzing the correlation between neurons and combining the application scenes,this research designs a fast fire-controlled modified Pulse Coupled Neural Network（FFC-MSPCNN）,which is achieved by further optimization and improvement of weight matrix W_ijkl,link strengthβand dynamic threshold amplitude V,with adding a variable parameter K to adjust the dynamic threshold.This model also simplify the setting parameters and retain the parameterαrelated to the threshold S’,with low computational complexity.Our proposed PCNN can acquire good breast mass segmentation results at only two iterations and has more biologically compatible and applicable expanded scenes.The accurate segmentation results of breast masses are post-processed by using morphological algorithms to obtain binary images of breast regions with smooth edges.The experimental results in MIAS and DDSM database show that our proposed method can accurately segment the breast masses and significantly reduce the randomness and unpredictability of neuronal discharge,which verifies the effectiveness and robustness of the method on the research.（2）Convolutional neural network based on deep learning can learn breast image features adaptively.According to the characteristics of breast images,through the learning of Conv Ne Xt network,we adopt Conv Ne Xt and introduce feed-forward convolutional neural network attention module CBAM to improve the learning ability of the network on the main features of breasts image.Additionally,we resort to the migration learning method to add weight data and improve network training efficient.This network can provide valuable information and evaluation criteria for physicians at the early detection and screening stage of breast diseases,and significantly reduces the workload of physicians.Our designed network is tested in the breast ultrasound database,MIAS database,and DDSM database,including 3600experimental images（2880 training images and 720 testing images）.We reasonably and effectively give evaluation results by using evaluation metrics and prediction data.The experimental results show that our propose method has high accuracy and low computational complexity and is an important application for the early diagnosis and auxiliary treatment of breast cancer lesions.