Detection of Oral Squamous Cell Carcinoma Using Mixup Augmentation and Deep Learning

Abstract

Oral Squamous Cell Carcinoma (OSCC) is a significant global health challenge where early and accurate detection is critical for improving patient outcomes. Histopathological examination is a method for diagnosing OSCC, but the issue is the labour, time spent and the observer variability. In this research, deep learning based OSCC identification from histopathological images was explored. Mixup data augmentation and Focal Loss to improve model outcomes. The study compared CNN architectures, EfficientNet, EfficientNetV2, MobileNetV3 and ResNet-50. Evaluation process included 5-fold stratified cross validation to get unbiased results. EfficientNetV2-Medium achieved the best overall performance, with an Accuracy of 0.9914, Precision of 0.9988, Recall of 0.9840, and an F1-score of 0.9913. These findings show that combining advanced data augmentation with modern CNNs can produce a robust and highly accurate tool to support pathologists in OSCC diagnosis.