Neural network-based segmentation model for breast cancer X-ray screening

Diagnostic efficiency of breast cancer screening remains one of the most important issues in oncology and radiology. Artificial intelligence technologies are widely used in clinical medicine to effectively solve a number of technological and diagnostic problems.
The aim. To develop segmentation neural network model for breast plain radiographs analysis with subsequent study of its clinical effectiveness.
Materials and methods. The artificial intelligence-based system was developed to analyze X-ray mammography, аnd included a segmentation neural network with the U-Net architecture, a classification neural architecture ResNet50 with outputting the result using gradient boosting. 15486 X-ray cases were used for training, estimation of diagnostic accuracy and validation of the developed segmental model. All cases were labeled in specially developed software environment LabelCMAITech. The segmentation accuracy was determined by Intersection over Union (IoU) similarity coefficient, the probability of malignancy was calculated using the binary classification metrics.
Results. The developed system is represented by a segmentation model based on neural network architecture. The model allows, with high accuracy of 0.8176 and higher, at threshold values on the output neurons of the network of 0.1 and 0.15, to localize X-ray findings that have diagnostic value for determining the likelihood of the presence of breast cancer signs in an X-ray mammographic study — focus, architecture distortion, local asymmetry, calcifications. When comparing the results of machine segmentation and marking of images by a radiologist, it was found that the model is not inferior to the doctor in the accuracy of determining the formations, extra-focal calcifications and intraglandular lymph nodes.
Conclusion. The results of this study allow considering the model as an intelligent assistant to a radiologist in the analysis of screening mammographic cases.

1. Siegel R.L., Miller K.D., Jemal A. Cancer statistics, 2019. CA A Cancer J Clin. 2019; 69: 7-34. https://doi.org/10.3322/caac.21551

2. Breast cancer. [web resource] URL https://www.who.int/cancer/prevention/diagnosis-screening/breast-cancer/en/ free (accessed September 16, 2020).

3. DeSantis C.E., Ma J., Gaudet M.M., Newman L.A., Miller K.D., Goding Sauer A., Jemal A., Siegel, R.L. Breast cancer statistics, 2019. CA A Cancer J Clin. 2019; 69: 438-451. https://doi.org/10.3322/caac.21583

4. Li Y., Chen H., Cao L., Ma J. A survey of computer-aided detection of breast cancer with mammography. J Health Med Inf. 2016; 4: 238. https://doi.org/10.4172/2157-7420.1000238

5. Welch H.G., Passow H.J. Quantifying the benefits and harms of screening mammography. JAMA Intern Med. 2014; 3: 448-454. https://doi.org/10.1001/jamainternmed.2013.13635

6. Curtis C., Frayne R., Fear E. Using X-Ray Mammograms to Assist in Microwave Breast Image Interpretation. Int. J. Biomed. Imag. 2012; 2012: 235380. https://doi.org/10.1155/2012/235380

7. Abdelhafiz D., Yang C., Ammar R. et al. Deep convolutional neural networks for mammography: advances, challenges and applications. BMC Bioinformatics. 2019; 20: 281 https://doi.org/10.1186/s12859-019-2823-4

8. American College of Radiology. The ACR breast imaging reporting and data system (BI-RADS) [web resource]. November 11, 2003. URL: http://www.Acr. org/departments/stand_accred/birads/contents.html. free (accessed February 27, 2020).

9. Ronneberger O., Fischer P., Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Arxiv. Lib. 2015; 15: 1-8. https://doi.org/10.1007/978-3-319-24574-4_28

10. He K., Zhang X., Ren S. Deep Residual Learning for Image Recognition. IEEE Comput. Vis. Pattern. Recognit. 2016; 10: 1-4. https://doi.org/10.1109/cvpr.2016.90

11. Natekin A., Knoll A. Gradient Boosting Machines, A Tutorial. Frontiers in Neurorobotics. 2013; 7: 21. https://doi.org/10.3389/fnbot.2013.00021

12. Wu J., Mahfouz M. R. Robust X-ray image segmentation by spectral clustering and active shape model. J Med. Imag. 2016; 3: 034005. https://doi.org/10.1117/1.jmi.3.3.034005