Abstract:
The early detection of oral cancer plays a pivotal role in enhancing the survivable rate of the patients. Recent advancement in artificial intelligence have made the diagnosis rapid and precise. The advent of deep learning has transformed medical image analysis, facilitating more precise, efficient, and automated evaluations of medical images. It serves the purpose of identifying and locating particular objects within medical imaging. The aim of this research is to develop a deep learning-powered system for diagnosing oral cancer, capable of distinguishing between cancerous and non-cancerous areas in a provided image. The Yolo is a cutting edge deep learning model employed for object detection, segmentation and classification. The system was retrained for the oral cancer dataset. The images are annotated with the help of the experts. A balanced dataset is created by data augmentation by rotating and flipping the images. The blurring is used to pre-process the images. The Yolov8 architecture has been enhanced through the integration of EfficientNet-B0 for the generation of feature maps, along with the implementation of a Feature Pyramid Network (FPN), which facilitates the detection of objects across various scales. Following that, the model is trained with the images and then validated using YOLOv8 model. The normal and abnormal part of an images are identified with a precision of 0.901. The mean Average Precision (mAP) obtained for the model is 0.913. The is YOLOv8 model is compared with other objection detection model such as YOLOv7, Mask R-CNN (Region based Convolutional Network) and Faster R-CNN. YOLOv8 is found to be the fastest object detection and classification framework compared to the other three models. These results greatly helps the medical practitioner to perform the initial investigation to and helps in the early detection of an oral cancer.
