Oral Lesions Classification Using EfficientNet Transfer Learning Model

Oral lesions are a significant diagnostic challenge due to their diverse range of pathologies. This research paper presents a study on automated classification of oral lesions using deep learning techniques, specifically the EfficientNetB7 model. The study uses the Oral Lesions: Cancer Detection Dataset, which comprises 2270 high-resolution images, to classify lesions into benign and malignant. The EfficientNetB7 architecture, known for its efficiency in handling large-scale image datasets, is used in the study. The model achieves an impressive accuracy rate of 99.12%, demonstrating its proficiency in distinguishing between benign and malignant lesions. The study also analyzes the model's performance, including sensitivity, specificity, and precision metrics, to emphasize its diagnostic reliability. The study also explores the interpretability of the model's predictions, highlighting key features contributing to its decision-making process. The research has implications beyond computer-aided diagnosis, enhancing the efficiency of oral lesion screening programs. The automated classification system holds promise as a supplementary tool for health workers, aiding in early and accurate identification of potentially malignant lesions, facilitating prompt intervention and improving patient outcomes.