Leveraging pretrained vision transformer for skeletal malocclusion classification using CLAHE and cut-out method

Skeletal malocclusion is a form of misalignment in the human jaw that can cause several issues with bones and teeth. Generally, there are three classifications of skeletal malocclusion. Utilizing deep learning techniques such as Vision Transformer (ViT). The classification of human jaw disorders, or skeletal malocclusion, helps in identifying abnormalities in the jaw so that accurate diagnoses and appropriate treatments can be administered by dentists to prevent ongoing pain from unnatural jaw alignment. Given the challenges posed by an imbalanced dataset, we employed Contrast Limited Adaptive Histogram Equalization (CLAHE) and cut-out techniques to enhance the quality of our input data. In this research, Pretrained Vision Transformer (ViT) is employed to classify three classes of skeletal malocclusion with relatively small data sets and simple preprocessing yet effectively classifying and predicting medical X-ray images. The results obtain an accuracy of 0.76 on the test dataset.