EffiSANet: EfficientNet Integration with Self-attention for Colorectal Cancer Classification

The highly fatal disease of colorectal cancer still has a substantial effect on world public health. Early detection of this malignancy is essential for extending human life and stopping its spread. Recently, image classification tasks have been highly influenced by CNN’s architecture because of its effective learning in hierarchical visual feature representation. However, CNNs are unable to recognize intricate patterns and variations in smaller datasets because they lack the diversity and volume of samples. Attention mechanisms in conjunction with CNNs can be utilized to alleviate these constraints. In this article, an efficient CNN-based model that is a combination of attention and an efficient net, denoted as the EfficientNet Self-Attention network (EffiSA-net), is proposed to classify colorectal cancer. Our experiment shows better results with the proposed model, achieving 95.49% test accuracy on the colorectal histology dataset. Our study offers a promising method for enhancing the effectiveness and precision of colorectal cancer categorization, which will eventually help patients by enabling earlier identification and intervention.