P03 Creating digital twin images of psoriasis using generative artificial intelligence

Abstract Introduction and aims Psoriasis is an inflammatory skin disorder affecting > 1 million people in the UK. Machine learning has become an important tool for image processing tasks within dermatology. An emerging technique within this field includes generative artificial intelligence (AI). The aim of this study was to train a generative adversarial network (GAN) to create personalized, AI digital twin images of psoriasis. Methods Overall, 1664 photographs of psoriasis and surrounding normal skin were taken from 57 participants. A total of 375 images from 20 individuals were used to train a GAN (StyleGAN) to generate ‘fake’ but realistic images of psoriasis. In total, 464 of these images were assessed and scored by three dermatologists who graded the images using the erythema and scaling components of the Psoriasis Area and Severity Index; mean erythema and scaling scores were 1.97 ± 0.011 and 1.71 ± 0.021, respectively, for images of moderate psoriasis, and 3.07 ± 0.018 and 3.00 ± 0.025, respectively, for images of severe psoriasis. Results In total, 210 real patient images were subsequently embedded into the latent space of the GAN allowing us to transform clinical photographs into digital twin images, replicating the unique psoriatic lesions of individual patients. Furthermore, comparison of images of different severity by the GAN enabled the identification of treatment and size vectors. Application of these vectors to the digital twin images allows us to observe a prediction for the changes that would be expected following treatment in the clinical setting, i.e. reduction in erythema and scaling with minimal reduction in size of psoriatic lesion. We are currently evaluating the accuracy of the personalized GAN images in predicting improvement in patients’ psoriasis undergoing treatment with ultraviolet. Conclusions This study shows that generative AI, using the StyleGan network, could be used to illustrate a predicted treatment response with antipsoriatic therapy for patients with psoriasis in a clinical setting.