A Comparative Study of CNN Architectures: ConvNeXt, MobileNetV3, and EfficientNet for Oral Disease Diagnosis

Oral diseases are a global health issue affecting billions of people worldwide. Early detection using artificial intelligence technology, particularly Convolutional Neural Networks (CNN), can enhance the accuracy of oral disease diagnosis. This study compares the performance of three CNN architectures—ConvNeXt Tiny, MobileNetV3 Large, and EfficientNet B0—in classifying oral diseases based on a medical image dataset. The evaluation considers accuracy, computational efficiency, and processing time. The results show that EfficientNet B0 achieved the highest accuracy (98.21%) with fewer parameters than ConvNeXt Tiny. However, ConvNeXt Tiny demonstrated the best absolute accuracy (96.96%) despite higher computational resource consumption. MobileNetV3 Large exhibited high efficiency in parameter usage with competitive accuracy (96.44%). Thus, the choice of CNN architecture depends on the trade-off between high accuracy and computational efficiency in clinical implementation.