Research On Intelligent Detection And Recognition Method Of Tumors Based On Breast Molybdenum Target Images

Breast cancer is the most common malignant tumor in women.Mammography is the most effective method in the diagnosis of breast cancer.However,due to the limited information that doctors can obtain from mammography images and the complex structure of the lesion area,even experienced experts may miss or misdetect.Therefore,it is proposed to use computer assisted doctors for breast cancer diagnosis to improve the accuracy of breast cancer diagnosis.In recent years,with the rapid development of computer technology,mass detection and recognition based on breast molybdenum target images has gradually become a research hotspot for computer-aided diagnosis(CAD)of breast cancer.This article focuses on the mass detection and recognition technology in computer-aided diagnosis of breast cancer.The main research contents are as follows:(1)In order to accurately detect the lump area(ROI)in the breast molybdenum target image,the working principle of Faster R-CNN algorithm is studied firstly,and the Caffe framework is built on Windows platform for training.Aiming at the phenomenon that the Faster R-CNN algorithm randomly selects samples by randomly setting the ratio of positive and negative sample parameters in random gradient descent small batch sampling,this paper proposes to introduce the Online Hard Example Mining(OHEM)technology to improve the Faster R-CNN algorithm.The specific method is:in the training process,calculate the loss of the extracted region of interest frame,and then sort the candidate loss regions according to the calculated loss value from high to low,and select the candidate region with poor current network performance as a difficult sample for standard ROI network training.The imbalance between positive and negative samples in the training process is solved.Compared with the above method,the discriminative ability of the algorithm is enhanced,and the detection accuracy of ROI is effectively improved.(2)This paper studies the benign/malignant method of breast masses based on support vector machine(SVM).Aiming at the problem that the classification performance of SVM is greatly affected by the penalty coefficient and kernel function parameter selection,a SVM parameter optimization method based on improved Fruit-fly optimization algorithm(LMFOA)is proposed.The LMFOA based SVM parameter optimization procedure is given.The LMFOA-SVM is applied to the benign/malignant identification of breast tumors.This paper not only compares the WDBC breast quantitative feature data set with LFOA-SVM algorithm,PSO-SVM algorithm,BP neural network,but also selects 120 DDSM breast image test set detection result pictures for mass segmentation and feature extraction.Based on the extracted quantitative features,breast benign/malignant identification experiments were performed.The results show that the LMFOA algorithm improves the classification performance of SVM and improves the accuracy of breast mass recognition.(3)Based on the above improved detection method and recognition algorithm,developed breast lump detection software and breast lump benign/malignant identification software on the MATLAB GUI platform.The breast lump detection software can quickly and accurately detect the number of breast lumps from DDSM breast images.And display location information.Breast benign/malignant identification software can give benign/malignant diagnosis results based on the extracted lump features.The software improves the diagnostic accuracy while reducing the workload of doctors,and greatly promotes the improvement of doctors’ work efficiency,and provides a feasibility study for digital intelligent medical care.