Computer-Aided Diagnosis of Oral Squamous Cell Carcinoma: A Feature-Based Transfer Learning Approach

Oral cancer, particularly Oral Squamous Cell Carcinoma (OSCC), has a high mortality rate due to late detection. However, manual diagnosis is difficult and time-consuming. Hence, the employment of machine learning methods has been explored to aid diagnosis through automated image classification. This study aims to evaluate pipelines combining pre-trained VGG19 convolutional neural network (CNN) model that is used to extract discriminative features from normal and cancerous oral histopathology images. The extracted features were fed to different machine learning models, support vector machine (SVM), k-nearest neighbours (kNN), and random forest (RF) were trained to classify the images. It was demonstrated that the VGG199-RF yielded the best performance across the training, validation, and test dataset with a classification accuracy of 99%, 92%, and 90%, respectively, against other pipelines evaluated. The study demonstrates that feature-based transfer learning is an attractive and effective approach to be employed for computer-aided diagnosis.