Prognosis Of Postoperative Tamoxifen Treatment For Breast Cancer Based On Automatic Mammographic Density Measurement

Breast cancer seriously threatens the lives and health of women around the world,tamoxifen therapy is often used in the adjuvant treatment of breast surgery,however,not all patients with good prognosis,which is not for all the postoperative patients with breast cancer could benefit from tamoxifen treatment,so the breast cancer tamoxifen treatment prognosis for patients with postoperative treatment and curative effect judgment has important significance.The problem in the prognosis of postoperative tamoxifen treatment for breast cancer is the lack of intuitively accessible biomarkers.In this paper,mammographic density change value(MDC),mammographic density change rate(MDCR)and other factors were proposed as prognostic imaging markers,and their prognostic effectiveness was studied.In order to obtain mammographic density(MD),this paper built an automatic measurement model of mammographic density,and proposed two kinds of networks with different data types to achieve automatic measurement of mammographic density by means of gland gray threshold regression.After obtaining MDC and MDCR,survival analysis was used to verify whether they could be used as prognostic imaging markers for postoperative tamoxifen treatment of breast cancer.For the automatic measurement model of mammographic density,the area of glandular tissue in breast image is calculated.In this paper,the gray threshold regression method of gland is proposed to obtain gland area,and two kinds of network structures are built to achieve gray threshold regression.The first one is Squeeze-and-Excitation Convolutional Neural Network(SE-CNN),which adopts a modular structure of multi-layer convolution to enhance the extraction ability of glandular features by deepening the network depth,and introduces channel attention mechanism to learn channel features with high dependence on results and great contribution in all channels,effectively improving the accuracy of results.The second is Graph Convolutional Network(GCN),where the input is a graph data structure that learns the information implicit between features.Firstly,construct graph data,extract texture,gray gradient and other features by observing image features and prior knowledge,take features as nodes,and set edges according to the correlation between features.Nodes and edges together constitute the graph structure.Then GCN was constructed,and the network architecture was composed of graph convolution layer,graph pooling layer,graph attention layer and full connection layer.The feature information of nodes in graph convolution layer was learned and updated,while the number of nodes in graph pooling layer kept decreasing,while the graph attention layer selectively retained nodes that could not be represented by other nodes.In this way,the node information passes through layers and finally outputs the grayscale threshold of the gland.Compared with the classical convolutional neural network Res Net50,the absolute threshold error of SE-CNN is 9.92±4.78 and the determination coefficient is 0.77,while the absolute threshold error of GCN is 11.32±4.29 and the determination coefficient is 0.71 and the absolute threshold error of Res Net50 is 10.94±5.29.The determination coefficient is 0.70.The results showed that SE-CNN performed better than other neural networks such as GCN and Res Net50 in breast image gland gray threshold regression task.For prognostic analysis,two prognostic imaging markers,MDC and MDCR,were proposed in this study,and their prognostic effectiveness was verified by survival analysis.When the subjects were grouped according to the scientific cutting value,the density map and restricted cubic spline map were drawn by R software to determine the cutting value,and the patients were grouped according to the cutting value.K-M analysis and Cox regression analysis were combined to determine the independent factors influencing the prognosis of postoperative tamoxifen treatment for breast cancer.HR of MDC =2.401,95% CI,1.031-5.587,P =0.165(>0.05),HR of MDCR =2.654,95% CI,1.102-6.395,P=0.030(<0.05),suggesting that MDCR was an independent risk factor for metastasis or recurrence of breast cancer treated with tamoxifen and had good prognostic ability.In this paper,the model of automatic mammographic density measurement based on SE-CNN is proposed to solve the problem of accurate calculation of mammographic density.In the prognostic analysis of postoperative tamoxifen treatment for breast cancer,the survival analysis results of MDCR meet the prognostic requirements,and can be used as a prognostic image marker of this treatment.