Oral squamous cell carcinoma, another name for oral cancer, is one of the ten most common diseases worldwide, accounting for approximately 500,000 new cases and 350,000 fatalities annually, with India accounting for one-third of these instances. There is an urgent need for objective, distinctive technologies that enable early, precise diagnosis. Through transfer learning on Inception-ResNet-V2, we built a method to identify images as “suspect” and “normal” and automated heat maps to highlight the area of the images most likely to be engaged in decision-making. Using the Kaggle datasets with photographic shots of 87 and 44 cases, we assessed the feasibility of the created approach. The system was tested using both 10-fold cross-validation and leave-one-patient-out validation procedure. The study’s novel findings and methods include developing and validating our styles on two datasets gathered from various locations in India, demonstrating that using patches rather than the full lesion image improves performance, and determining which areas of the images are predictive of the classes using class activation map analysis.