Klasifikasi Citra Jenis Lesi Lidah dalam Antisipasi Dini Kanker Mulut dengan EfficientnetV2

Abstract

Oral cancer is one of the most dangerous types of cancer and is often diagnosed late. Tongue lesions are frequently an early sign of this condition, making early detection of tongue lesions crucial to improving patient recovery. Tongue lesion diagnosis is typically performed using dermoscopy or histopathology methods, which require biopsies and are time-consuming and costly. This study focuses on classifying tongue lesions into four classes: Normal, Benign, OPMD, and OSCC. The architecture used in this study is EfficientNetV2 for feature extraction and model training. The EfficientNetV2 architecture employed in this research is an advancement of EfficientNet, which has proven to be efficient and effective in image processing, particularly in image classification tasks. This architecture incorporates various modifications to enhance model efficiency and accuracy in image processing. The study utilizes 1000 image data samples, divided into training, validation, and testing datasets. The results of using the EfficientNetV2 architecture show an accuracy of 96.88% with 50 epochs and a batch size of 32, using a data split ratio of 80% training data, 10% validation data, and 10% testing data. The findings of this research conclude that the architecture used is capable of classifying the four types of tongue lesions effectively.