A Pre-trained MobileNetV2 Model for Oral Cancer Classification

In emerging economies and regions with low to middle incomes, oral cancer has emerged as one of the prevalent malignant tumours and a highly critical medical condition. Timely detection and appropriate treatment can often avert fatality for the majority of cancer cases. Conversely, delayed diagnosis and delayed treatment initiation can pose life-threatening risks. Hence, early identification of oral cancer plays a pivotal role in saving lives. Early detection coupled with effective treatment of oral cancer can substantially lower the mortality rate associated with the disease. Future advancements in cancer detection and diagnosis could witness a transformative impact through the integration of developments in biomedical optics and deep learning techniques. In the proposed work, oral cancer prediction has been derived by harnessing data from Kaggle and utilizing a Convolutional Neural Network-based pre-trained MobileNetV2 model. This model has been fine-tuned with different batch sizes and epoch values. Notably, employing a batch size of 64 and training for 5 epochs yielded an accuracy of 66.80% and a loss of 0.6131. However, by modifying the batch size to 128 and epoch count to 10, the accuracy improved while the loss diminished. The values of accuracy and loss have been identified as 78.12% and 0.5138, respectively. Further, in future, to achieve a better performance outcome, the model can be implemented by increasing the epoch values.