Benign Dermoscopic Morphology by Skin Tone and Anatomic Site

The morphologies of benign lesions are well-defined in the literature; however, most visual examples and training materials are selectively sampled from patients with lighter skin and in the most common anatomical areas, leaving important gaps in how common, harmless lesions appear in darker skin tones. As a result, both clinicians and AI tools can struggle to identify benign lesions accurately in all patients, leading to over-biopsying of patients with skin tones in the extremes of the spectrum or patients who present with benign lesions in uncommon anatomical areas.

Our project will create an inclusive, high-quality annotated image dataset of common benign skin lesions, such as seborrheic keratoses, angiomas, and dermatofibromas, captured across the full range of objectively measured skin tones and body sites. Using dermoscopic imaging and a color-measurement tool that quantifies skin tone precisely (via Individual Typology Angle, or ITA), we will document how these benign lesions vary in dermoscopic appearance. We will assess the innate skin tone of each patient by measuring the ITA value of the inner portion of the upper arm to avoid incorporating variations in skin tone due to UV exposure on different body sites. This dataset will improve the ability of clinicians to recognize benign skin lesions more accurately and reduce unnecessary biopsies. By capturing how benign lesions look across diverse innate skin tones, this work will improve clinical training and accuracy of melanoma diagnosis and help ensure that future AI tools are built on datasets that are representative of a more diverse population of patients.