OralCANet: Transformer-Based Oral Cancer Detection using Q-Layered Classification

Oral cancer is a problem for people all, around the world. The thing is, doctors need to find cancer and bad cells on so people can get better. Usually, doctors look at the mouth. Take a sample of the bad cells. This way of doing things is not very good because it is based on what the doctor thinks and it takes a long time. Sometimes doctors make mistakes. Deep Learning is a way of doing things that seems to be working. The old models that use Deep Learning have some problems. They get confused by information and they have trouble understanding complicated things in pictures of sick cells. Oral cancer and the pictures of cells are what make this hard for Deep Learning to figure out. This paper is trying to fix some problems so it suggests something called OralCANet. This is a kind of system that has many layers and it does a lot of things to make the information it gets better. It uses something called ResNet-101 that it already knows and it adds some things, like Wavelet Transform and Butterworth Filtering to make the information clearer. To find the parts it uses a few different methods together like DeepLabV3+, GrabCut and Mask R-CNN to make sure it gets the right areas. The main thing that makes OralCANet special is the Q-Layered classification part. This part is unique because it uses a few things together. It uses a Transformer Encoder to look at the picture and find features that are important everywhere. It also uses a BiLSTM to look at how thingsre connected in a sequence. It uses an Attention-based Deep Feature Encoding Layer to focus on the features that are bad. The people who made OralCANet tested it on two sets of pictures. One set was of tissue samples. The other set was of mouth pictures. OralCANet did a job and was right 99 percent of the time. It even did better, than models that are currently the best. These results suggest that the Q-Layered integration offers a robust, high-precision tool for automated oral cancer screening, potentially assisting clinicians in making faster, more accurate diagnoses.