A Novel Method to Detect Oral Carcinoma Using Box Annotation Based on YOLO Model

Oral carcinoma affects millions of individuals globally and is a major public health problem due to its high fatality rate. Treatment effectiveness depends on early detection and improves survival rates. An effective method for automated oral cancer screening is deep learning. The paper's primary goal is to use deep learning techniques to advance the state of oral cancer diagnosis, specifically YOLOv5 and YOLOv7. We aimed to achieve accurate detection by leveraging bounding box annotations for both two distinct types of data: collected clinical data and benchmark data of oral carcinoma. The proposed methodology encompasses data collection, preprocessing, expert-driven bounding box annotation and the use of cutting-edge deep learning models for lesion identification, YOLOv5 and YOLOv7. Bounding box serving as expert annotations are meticulously added to pinpoint cancerous regions within the images. Subsequently, a comprehensive evaluation is conducted, measuring the performance of both models based on precision, recall rate (sensitivity) and mean average precision (mAP). Our findings reveal that YOLOv5 outperforms YOLOv7 in terms of oral cancer detection, demonstrating superior precision and recall rates. In the evaluation of this research, the YOLOv5 model demonstrated strong performance metrics on both clinical and lip and tongue cancer datasets. For the lip and tongue cancer dataset, the YOLOv5 model achieved a precision of 97.2%, a mAP (mean Average Precision) of 97.7% and a recall of 96.3%.