Tumor cell invasion and metastasis is closely related to various components in tumor microenvironment, and the spatial distribution of tumor cells may be linked to the tumor progression. The directionally distributed collagen fibers would lead to one-way migration of tumor cells, and misarranged collagen fibers allow tumor cells to multi-directional migration in tumor boundary, and therefore tumor cells present in different spatial distributions in the microenvironment. In this study, the open source convolutional neural network Hover-Net was used for segmentation and classification. After accurately segmenting tumor cells, spatial features were extracted. Our results indicated that there were significant differences in the spatial distribution of tumor cells in the two modes. These were further demonstrated to be potential different prognostic patterns.
|