We propose a deep learning based digital pathology method that can classify colorectal cancers from digitized whole slide images. The conventional digital pathology methods approach cancer grading as a categorical classification problem, where the goa...
We propose a deep learning based digital pathology method that can classify colorectal cancers from digitized whole slide images. The conventional digital pathology methods approach cancer grading as a categorical classification problem, where the goal is to classify them into appropriate classes. However, in the case of cancer cells, the higher the grade or differentiation of each class, the poorer the condition of the cancer is, making simple categorical classification insufficient to address this issue. Therefore, in this paper, we formulate cancer grading as both categorical and ordinal classification problems and conduct two cancer grading tasks simultaneously. To achieve this, we build a deep learning model based on vision transformer and order learning. The proposed method is evaluated using a colorectal tissue dataset. Experimental results show that our method is able to accurately classify cancer grades and outperforms other competing models.