Neural network combining X-ray and ultrasound in breast examination

• Published in: Neural computing & applications Vol. 34; no. 5; pp. 3523 - 3535
• Main Authors: Song, Jiaguang, Zhang, Yuezhong, Wang, Shi, Liu, Zhi, Sun, Dianmin
• Format: Journal Article
• Language: English
• Published: London Springer London 01.03.2022; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Artificial neural networks; Classification; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Error analysis; Extreme values; Image Processing and Computer Vision; Image segmentation; Mammography; Muscles; Neural networks; Performance evaluation; Probability and Statistics in Computer Science; Special Issue on Multi-modal Information Learning and Analytics on Big Data; Training; Ultrasonic imaging; Ultrasound examination; Neural network; AdaBoost integrated neural network; X-ray examination
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-021-05882-2

With the acceleration of people’s life rhythm, the incidence of breast cancer is gradually increasing. This study mainly explores the application of neural network combined with X-ray and ultrasound in breast examination. The process discussed in this research is mainly based on AdaBoost integrated neural network. According to the basic principles of artificial neural networks, first use training samples to train the neural network, then form an integrated neural network based on the AdaBoost algorithm, and then use the test samples to perform prediction tests on the network, and perform calculations in the integrated neural network to predict the classification the results are compared with the actual results in the test set, and the total number of misclassifications is analyzed to determine whether the prediction effect of the integrated neural network is good or bad. In the process of training the neural network, the analysis of the obtained case data needs to be numerically processed and then normalized to map the data to the [− 1,1] interval. The segmentation goal is to segment each tissue in the breast ultrasound image, which are skin layer, fat layer, gland layer, tumor, muscle layer, and background layer. The mean squared error is used as the network performance evaluation function. The research results are measured by the error rate. The lower the error rate, the better the classification prediction effect and the better the performance of the individual neural network. The diagnostic coincidence rate combined with X-ray and ultrasound was 88.33% (265/300). Compared with pure ultrasound or mammography, the difference was significant ( P  < 0.05). The recall rate of neural network combined with X-ray and ultrasound reached 91.4%. The results show that the neural network combined with X-ray and ultrasound has extremely high application value in breast examination.