The Histopathologic Oral Cancer Detection Dataset, which consists of meticulously annotated high-resolution histopathological images, is an essential resource for advancing the early diagnosis and classification of oral cancer. The dataset, categorized into "Normal" and "Oral Squamous Cell Carcinoma (OSCC)" classes, underpins the development and evaluation of sophisticated deep learning models, particularly Convolutional Neural Networks (CNNs), designed to distinguish between malignant and non-malignant tissue samples. In this study, the efficacy of the ResNet50 deep learning architecture was rigorously evaluated for its ability to classify histopathological images of oral cancer. Two methodologies were investigated: initially, ResNet50 was applied as an independent classifier, achieving a Precision, Recall, F1 Score, and Accuracy of 97.43%, alongside an MCC of 94.86 and an AUC of 97.43%. Subsequently, the study incorporated Contrast Limited Adaptive Histogram Equalization (CLAHE) during the pre-processing phase, a technique well-regarded for enhancing image contrast adaptively, particularly in medical imaging contexts. The integration of CLAHE resulted in a marked improvement in performance, with the model attaining a Precision of 98.37%, Recall of 98.33%, F1 Score of 98.33%, Accuracy of 98.33%, MCC of 96.70%, and AUC of 98.37%. These results emphasized the importance of CLAHE. It is very useful for early diagnosis of oral cancer.