Mammographic Mass Detection Based on Data Separated Ensemble Convolution Neural Network

The number of female deaths caused by breast cancer each year is the second of all causes of death. According to the International Agency for Cancer's Research, early diagnosis can effectively reduce breast cancer mortality. Therefore, our study attempts to construct an automatic X-ray image detection system to help radiologists Interpret the image content to improve the accuracy of diagnosis. In recent years, the development goal of convolutional neural networks is to achieve relatively good accuracy with the least amount of calculation to meet the actual application situation, but ignore the application fields that do not emphasize real-time calculation. Our research will try to use ensemble learning methods to integrate several Efficientnets through two integrated strategies to improve the accuracy of the model. Our study used the DDSM data set, which contains 1329 cases and 5316 digital X-ray images. 641 patients were diagnosed as negative and 688 were diagnosed with benign or malignant tumors. Experimental results show that the accuracy and recall rate of the model can be improved through ensemble learning. The framework proposed in this experiment also achieves an accuracy rate of 87.6 and a recall rate of 91.8%.