Dermatological imaging of skin lesions in different skin phototypes using style transfer

In dermatology, accurate identification and classification of skin lesions in various skin phototypes are essential for early and effective diagnosis. In this context, advances in artificial intelligence have highlighted the potential of Convolutional Neural Networks as powerful tools for generating realistic medical images. This work aims to improve visual data availability to support computer-aided diagnosis. Beyond extending a dataset only with data augmentation techniques, we seek to enrich the quality and diversity of the data set including images generated for each proposed skin phototype. This study focuses specifically on the generation of dermatological images of skin lesions in six skin phototypes (Fitzpatrick scale), using the CNN model (VGG19) and applying Style Transfer. Once the images are generated, they are evaluated using two metrics: the structural similarity index and the Kullback-Leibler divergence. Seven hundred clinical images of seven common skin lesions were collected using the HAM10000 dataset, and 42 images were generated with the VGG19 model. The results suggest that applying the ST technique as a new approach to diversifying skin phototypes produces good results.