Tumor segmentation in whole-slide histology images using deep learning
Abstract
The paper addresses the problem of segmentation of malignant tumors in large whole-slide histology images in the context of computer-assisted diagnosis of breast cancer. The method presented in this study is based on image classification procedure of norm/tumor type. The procedure calculates probability of belonging of each particular elementary image region of 256×256 pixels to the "tumor" class, which are isolated by corresponding sliding-window technique. The procedure capitalizes on convolutional neural networks and Deep Learning methods. The neural networks being employed were trained on a representative dataset of 600 000 fragments sampled from whole slide images and representing the morphological and colorimetric variability of two classes. The resultant probability maps were post-processed using conventional image processing algorithms to obtain the final binary masks of pathological regions. The proposed algorithm of segmentation of whole slide histological images can be used in computerized diagnosis of cancer for detection and segmentation of malignant tumors.
Keywords
whole-slide histology images,
deep learning,
convolutional neural networks,
image segmentation,
computerized cancer diagnosis
Supporting agencies: The work was done with partial financial support in the framework of the presented grant of the President of the Republic of Belarus (by order of the President of the Republic of Belarus no. 32rp dated January 19, 2018), as well as according to the BRFFR project "Development of high-performance means of intellectual software support for solving labor-intensive informatics and bioinformatics problems" (under the contract number F18R-178 dated May 30, 2018).
About the Authors
V. A. Kovalev
The United Institute of Informatics Problems of the National Academy of Sciences of Belarus
Belarus
Vassili A. Kovalev, Cand. Sci. (Eng.), Head of the Laboratory of Biomedical Images Analysis
Belarus
Vassili A. Kovalev, Cand. Sci. (Eng.), Head of the Laboratory of Biomedical Images Analysis
V. A. Liauchuk
The United Institute of Informatics Problems of the National Academy of Sciences of Belarus
Belarus
Vitali A. Liauchuk, Researcher
Belarus
Vitali A. Liauchuk, Researcher
A. A. Kalinovski
The United Institute of Informatics Problems of the National Academy of Sciences of Belarus
Belarus
Alexander A. Kalinovski, Researcher
Belarus
Alexander A. Kalinovski, Researcher
M. V. Fridman
Minsk City Clinical Oncologic Dispensary
Belarus
Mikhail V. Fridman, Dr. Sci. (Med.), Head of the Pathoanatomical Laboratory
Belarus
Mikhail V. Fridman, Dr. Sci. (Med.), Head of the Pathoanatomical Laboratory
References
1. Danilov R., Borovaja T. Gistologija, jembriologija, citologija. Histology, Embryology, Cytology. Moscow, GJeOTAR-Media, 2018, 520 р. (in Russian).
2. Luo W., Hassell L. A. Use cases for digital pathology. Digital Pathology: Historical Perspectives, Current Concepts and Future Applications. Springer International Publishing, 2016, pр. 5–15.
3. Beckwith B. A. Standards for digital pathology and whole slide imaging. Digital Pathology: Historical Perspectives, Current Concepts and Future Applications. Springer International Publishing, 2016, рр. 87–97.
4. Saco A., Bombi J. A., Garcia A., Ramírez J., Ordi J. Current status of whole-slide imaging in education. Pathobiology, 2016, vol. 83, pp. 79–88.
5. Szegedy C., Liu W., Jia Y., Sermanet P., Reed S., …, Rabinovich A. Going deeper with convolutions. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2015). Boston, 2015, pp. 1–9.
Review
For citations:
Kovalev V.A., Liauchuk V.A., Kalinovski A.A., Fridman M.V. Tumor segmentation in whole-slide histology images using deep learning. Informatics. 2019;16(2):18-26. (In Russ.)
Views: 854
Email this article 