Automatic Detection of Butterfly Malar Rash in Facial Images Using Deep Learning

Lupus is an autoimmune disorder with the majority of the cases being Systemic Lupus Erythematosus (SLE). Diagnosis of SLE is time consuming due to the broad range of tests, with no one test able to give a definitive diagnosis by itself. A common indicator for SLE is a butterfly wing shaped rash across cheek and nose area. In this study, we propose a deep learning model to identify butterfly malar rash (BMR) from facial images. We construct a dataset of facial images containing BMR and non-BMR rash patterns by collecting images from social media and web search. For experimentation, we used the pre-trained models DenseNet-121, ResNet50, VGG16, and Xception, and then fine-tuned these models on our dataset. Our experimentation shows that among the considered CNN models DenseNet-121 achieves the highest accuracy of 90.98 % in detecting BMR in facial images.