YOLOv7 for Dental Caries Detection: Challenges in Small Object Identification

In countries like India, dentistry is often a neglected healthcare sector. Due to several factors such as poor eating habits, poverty, socio-economic challenges, illiteracy, and a lack of awareness, dentistry tends to take a backseat compared to other health fields. However, small lesions on teeth can lead to serious problems in the future. To address this issue, an "automated dental caries detection system" is proposed. A digital database of RGB images was developed using smartphones, comprising photos from 194 patients. Each patient contributed 5 to 6 images to ensure comprehensive coverage of the entire oral cavity. Under the guidance of a dental practitioner, these images were labeled to indicate dental caries stages and identify missing or fractured teeth (which were excluded from the study) using the Azure Custom Vision model. The images were then converted to the YOLO (You Only Look Once) format. YOLOv7 was implemented to detect the location of "dental caries". Although YOLOv7 performs well in object detection, it struggles to deliver satisfactory results with smaller objects, showing only 62% precision. While YOLOv7 excels in real-time object detection, it struggles with identifying smaller objects in dental images, leading to unsatisfactory precision.