ColNet: A Compact Convolutional Transformer Framework for Automated Colon Cancer Prediction from Endoscopy Images

Colon cancer is one of the most causes of cancer-related deaths worldwide. Recent developments in deep learning introduced automated diagnostic systems, with Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) playing a significant role in cancer detection. However, they struggle with capturing long range dependencies, while ViTs require large scale datasets, making them less suitable for medical applications with limited data availability. Compact Convolutional Transformers (CCTs) were introduced as a hybrid deep learning approach, combining the local feature extraction of CNNs with the global self-attention mechanisms of Transformers. This paper shows implementation of a Compact Convolutional Transformer model for detecting colon cancer from images. It consists of a series of steps including input layer definition, data augmentation, tokenization of image patches, and Transformer Encoder Layers and performance evaluation metrics are discussed. Proposed model provides an accuracy score of 97.54%, precision score of 97.24%. Also shown, how it is more precise than traditional deep learning models such as CNN, ResNet and vision transformer, efficient and accurate. This study marks a significant advancement in colon cancer prediction, offering promising avenues for better diagnosis and improved patient outcomes.