Automatic Botox Injection Points Detection Based on Deep Learning

Cosmetic dermatology, with Botox injections being one of the most popular treatments, relies heavily on accurately identifying injection points, which requires expert evaluation of wrinkles and expression lines. This manual process is time-consuming, labor-intensive, and prone to variation among experts. Given the complexity of facial anatomy, there is a clear need for an automated detection system to improve the precision and efficiency of Botox injection point identification. In this paper, we applied YOLOv5, YOLOv7, and YOLOv8 methods for detecting Botox injection points. The best results, surpassing 70% accuracy, were achieved using YOLOv8 across various facial expressions. This indicates that YOLOv8 is particularly effective in handling the challenges associated with detecting small and subtle features in facial expressions, demonstrating its superiority over the other models for this task. These results highlight the potential of YOLOv8 to enhance the precision and reliability of automated Botox injection point detection in clinical settings. Additionally, we proposed a novel dataset consisting of 1,948 clinical facial images collected from 487 patients, which serves as the first comprehensive collection dedicated to Botox injection point detection, providing a unique resource for future research in this area.