Real-world use of a deep convolutional neural network to assist in the diagnosis of pyoderma gangrenosum

Early diagnosis of pyoderma gangrenosum (PG) can be challenging. A delayed or inaccurate diagnosis can lead to significant morbidity. PG can be misdiagnosed as cellulitis, ecthyma, or other infectious disease processes, leading to unnecessary debridement or other surgeries that can significantly worsen PG. There has been an abundance of research using artificial intelligence-powered machine learning and deep learning to create models to assist in dermatologic diagnosis. However, these tools often remain embedded in research institutions with limited use in real-world settings.1 A machine learning algorithm to aid in distinguishing PG from venous ulcers based on clinical images alone was recently developed, ultimately reaching a sensitivity of 97%