DermDiff: A Diffusion-based Multitasking Dermatology AI for Bias-Free Diagnosis (RPA Pilot/ Seed Project)

Skin diseases, such as skin cancer, are a significant public health issue, and early diagnosis is crucial for effective treatment. However, access to dermatologists, particularly for marginalized communities, can be limited, leading to delays in diagnosis. Artificial intelligence (AI) algorithms have the potential to assist in triaging benign versus malignant skin lesions and improve diagnostic accuracy. However, current AI models for skin disease diagnosis are often developed and tested on limited and biased datasets, leading to poor performance on certain skin tones and ethnicities. To address this problem, we propose to develop a multitasking deep learning model, namely DermDiff, that leverages a diffusion model to generate diverse and representative data for skin disease diagnosis. The model will also incorporate fairness constraints to ensure that the model performs equally well across different skin tones and ethnicities. In addition to classifying skin lesions as benign or malignant, the model will also perform lesion segmentation to assist in diagnostic decision-making. To evaluate the proposed model, we will use a combination of publicly available datasets and generated images of diverse skin tones. The performance of DermDiff will be compared to existing state-of-the-art models for image generation capability as well as for skin disease diagnosis. The generated diverse and representative training data could also be used to improve the performance of other AI models for skin-related tasks. The successful completion of this project should lead to the development of a clinically useful tool for the early detection and diagnosis of skin diseases, particularly for marginalized communities.