Dental lesion detection in medical imaging remains a critical challenge in diagnostic dentistry. This research leverages a deep learning-based detection approach for early-stage lesion identification. A custom-curated dataset comprising 242 images, both high-quality healthy and periapical lesion-based radiographs, was utilized. After going through all the techniques, the horizontal flip was used as a data augmentation technique to increase the size of the dataset. The study evaluates multiple CNN architectures such as AlexNet, InceptionV3, and DenseNet-121 on periapical images. Extensive experiments, including fivefold cross-validation and independent test evaluations, demonstrate DenseNet-121 achieving the highest accuracy at 95.83%, followed by InceptionV3 and ALexNet. Challenges addressed include class imbalance and dataset variations, while insights from comparative model performance highlight DenseNet-121 as optimal for robust lesion detection. The results emphasize the transformative potential of AI in enhancing diagnostic accuracy and accessibility, setting a foundation for advanced dental imaging applications using deep learning.