A Novel Attention-Based ConvMixer Model for Enteroscope Biopsy Histopathological Images

Histopathological image analysis is the gold standard used by pathologists to diagnose various forms of colorectal cancer. However, pathologists’ diagnoses are highly subjective and prone to misdiagnosis. Therefore, it is needed to develop computer-aided methods to classify colorectal images since they may eliminate inter-operator variability inherent in personal evaluations. In this study, an effective approach for enteroscope biopsy histopathological image classification based on ConvMixer networks is presented. To further improve the performance of the proposed network, a channel-based attention layer has been integrated. To the best of our knowledge, this is the first study that exploits a channel attention based on ConvMixer approach for colon cancer histopathological image classification. The experimental results conducted on the recently released Enteroscope Biopsy Histopathological H&E Image Dataset (EBHI), confirm that the proposed method has promising capabilities compared to the state-of-the-art.