<![CDATA[Botulinum Toxin (BOTOX) injections are widely used as a non-invasive cosmetic treatment to enhance facial appearance. However, determining the optimal dosage still relies on subjective visual assessment by medical professionals, which can lead to inconsistency. This study proposes a model deep learning–based classification framework using convolutional neural networks (CNNs) to automate BOTOX dosage prediction from forehead wrinkle images. Four CNN architectures Inception-V3, ResNet-34, ResNet101-V2, and EfficientNetB2 were evaluated on an augmented dataset of 168 cropped images, equally divided between 2-unit and 4-unit dosages. The dataset was generated through flipping and rotation augmentation to address class imbalance and enhance model generalization. Among the models, EfficientNetB2 achieved the highest accuracy of 92.8%, surpassing Inception-V3 85.7%, ResNet-34 82.1%, and ResNet101-V2 80.3%. The superior performance of EfficientNetB2 reflects its capability to extract fine-grained wrinkle patterns efficiently while maintaining computational efficiency. The novelty of this work lies in integrating CNN-based visual feature extraction with expert-labeled clinical image data for objective BOTOX dosage determination. Although limited by dataset size, this study highlights the potential clinical application of deep learning in supporting accurate, consistent, and data-driven facial aesthetic treatments.]]>
Copyrights © 2025
