Multi-scale fully convolutional network for gland segmentation using three-class classification

• Published in: Neurocomputing (Amsterdam) 2020-03, Vol.380, p.150-161
• Main Authors: Ding, Huijun, Pan, Zhanpeng, Cen, Qian, Li, Yang, Chen, Shifeng
• Format: Article
• Language: English
• Subjects: Dilated convolution; Fully convolutional network; Histological image; Multi-scale; Segmentation
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2019.10.097

Automated precise segmentation of glands from the histological images plays an important role in glandular morphology analysis, which is a crucial criterion for cancer grading and planning of treatment. However, it is non-trivial due to the diverse shapes of the glands under different histological grades and the presence of tightly connected glands. In this paper, a novel multi-scale fully convolutional network with three class classification (TCC-MSFCN) is proposed to achieve gland segmentation. The multi-scale structure can extract different receptive field features corresponding to multi-size objects. However, the max-pooling in the convolution neural network will cause the loss of global information. To compensate for this loss, a special branch called high-resolution branch in our framework is designed. Besides, for effectively separating the close glands, a three-class classification with additional consideration of edge pixels is applied instead of the conventional binary classification. Finally, the proposed method is evaluated on Warwick-QU dataset and CRAG dataset with three reliable evaluation metrics, which are applied to our method and other popular methods. Experimental results show that the proposed method achieves the-state-of-the-art performance. Discussion and conclusion are presented afterwards.