Cancer is a life-threatening disease represented as uncontrolled and abnormal growth of cells. The survival chances can be increased, if cancer disease detected in early stages. Automated diagnostic models provide a quicker and more adaptable alternative to physical examination. This study proposes a deep learning-based model for multi-class cancer classification using medical images. A CNN backbone namely ResNet-50 is utilized to capture spatial features from input images. These features are converted into graph structures, such that each image patch treated as a node. Later, the Attention-Enhanced Graph Neural Networks (AE-GNNs) process generated graph structure to obtain complex spatial relationships, which plays vital role in multi-class cancer classification. The performance of proposed model was evaluated using standard performance metrics and compared with base models like VGG16, LMHistNet, and BiLSTM. The results generated by experiments done by the proposed model exhibits as overall accuracy of 91.91%, highlighting to be utilized in automated cancer diagnosis.