<p>Oral cancer, particularly oral squamous cell carcinoma (OSCC), is a severe condition characterized by complex etiologies and a high mortality rate. Early detection and accurate diagnosis are critical for improving patient outcomes and reducing morbidity. Traditional diagnostic techniques, such as histopathological examination, are time-consuming, susceptible to human error, and highly reliant on the expertise of pathologists. Recent advances in deep learning, particularly convolutional neural networks (CNNs), have facilitated the development of computer-aided diagnostic systems for cancer detection. However, conventional CNNs often struggle to effectively identify the most informative features in histopathological images, limiting their ability to capture subtle and complex cancerous patterns. To overcome this limitation, we propose a fine-tuned DenseNet-121 model augmented with a customized attention block that includes parallel spatial and channel attention modules. The spatial attention module guides the model to focus on the most relevant regions of the image, while the channel attention module emphasizes the most important channels. This attention block is embedded at multiple levels of the DenseNet-121 architecture, enhancing feature extraction and ultimately improving diagnostic accuracy. The proposed model was evaluated on two diverse histopathology datasets, achieving state-of-the-art results. The model achieved 97.88% and 93.50% accuracy on the respective datasets without augmentation, which improved to 99.23% and 95.51% with augmentation.</p>