Skin Cancer Detection and Tracking using Data Synthesis and Deep Learning

Dermatology is a medical field which stands to be heavily augmented by the use of artificial intelligence techniques. Diseases are visually screened for, and many disease diagnoses are performed strictly with an in-clinic visual examination. Discerning between skin lesions is difficult the difference between skin cancer (melanoma, carcinoma) and benign lesions (nevi, seborrheic keratosis) is minute (Figure 1). With 5.4 million cases of skin cancer diagnosed each year in the United States alone, the need for quick and effective clinical screenings is rising [1, 2]. Patients with skin cancer tend to be afflicted with many moles, and so one of the challenges in skin cancer screenings is identifying them amongst a myriad of benign lesions. Another key element of these diagnoses is based on inspecting temporal changes in lesions a fast changing lesion is more likely to be malignant. As such, patients and providers need tools to support this at scale.