The catastrophic impacts of the lethal illness known as intestinal cancer are felt by those who suffer from it. Colonoscopy is the most effective cancer screening procedure currently available. While several deep learning methods are already used by existing systems, little research has been conducted on feature extraction and testing in the context of cancer diagnosis. The suggested system is a more comprehensive implementation of CADx technology that relies on intelligent detection. Data from the CKHK-22 dataset was preprocessed and kept in the cloud before being utilized for the system's training and testing purposes. To create local binary pattern (LBP) features, we first grayscale the colour features included in the colonoscopy dataset images. Several convolutional neural networks (CNNs), including VGG-16, DenseNet-201, and ResNet-50, are trained on the CKHK-22 dataset after it has been transformed to three features. In a vital step, the CKHK-22's feature acquisitions are fused to produce new features that are then employed in the DV-22, RD-22, and RDV-22 fusion CNN models, respectively. The CADx system, transfer learning with LSTM, and DenseNet-201 all attained testing accuracy rates of 91.92%, 86.76%, and 68.56%, respectively, over the original, grayscale, and LBP image feature datasets. By initially collecting the combined features of the original, grey, and LBP features through transfer learning with LSTM, the RDV-22 model obtained the best testing accuracy (90.81%) while applying fusion CNN for feature acquisition. The CKHK-22 dataset was mined for its three most distinguishing features by this CADx system, which yielded sufficient accuracy with DenseNet-201 and merged features and optimal accuracy with RDV-22 fusion CNN in diagnosing cancer. Using feature extraction, cancer diagnoses may be predicted with high precision.