Transformer Gated Fusion Network For Predicting Lung And Colon Cancer

Abstract

Cancer remains one of the leading causes of mortality worldwide. Recent studies have demonstrated that hybrid deep learning models can effectively detect lung and colon cancers by analysing complex histopathological images, highlighting their potential to support accurate and efficient cancer diagnosis. The system integrates machine learning with deep learning using ResNet18 to extract image features, which then interact with classifiers and transformer networks for information fusion. Integrating XGBoost with ResNet18 achieved high classification accuracy, followed by deep learning-based information fusion approaches involving CNNs and Transformers. The highest accuracy was obtained by integrating wavelet-based feature extraction with CNN and Transformer architectures, outperforming the standalone deep learning models. These findings highlight the promising potential of hybrid and multimodal approaches for medical diagnosis. Further improvements in these techniques are essential to enhance diagnostic accuracy and reduce the likelihood of errors in clinical decision-making.