Skin Type Determination Using Image Artificial Intelligence

Skin pigmentation, skin phototype, and photodamage are important determinants of skin cancer risk, dermatological disease presentation, treatment response, and prognosis. They also influence the performance of artificial intelligence (AI) systems developed for dermatological diagnosis and decision support. Despite their clinical importance, these characteristics are commonly assessed using subjective classification systems with limited accuracy and reproducibility. Misclassification occurs both in self-reported and clinician-reported assessments, which may reduce the precision of individualized risk assessments, prevention strategies, and clinical decision-making. Current classification methods often rely on the Fitzpatrick skin phototype scale and visual assessment of skin pigmentation and photodamage. Although widely used, these approaches have recognized limitations, particularly across diverse populations and skin tones. Objective measurement techniques, such as reflectance spectrophotometry and colorimetry, provide more accurate assessments of skin pigmentation but are not routinely available in clinical practice and do not directly measure phototype or accumulated photodamage. Consequently, there is a need for more robust, scalable, and objective methods to characterize skin pigmentation, phototype, and photodamage. Recent advances in deep learning have demonstrated high performance in image-based medical applications, including dermatology. However, most dermatological AI systems have focused on lesion detection and classification, while AI-based assessment of fundamental skin characteristics remains underdeveloped. Furthermore, many existing AI systems have been trained on datasets lacking detailed and reliable information on skin pigmentation, phototype, and photodamage, limiting their generalizability and raising concerns regarding fairness and performance across different skin types. This international multicenter observational study aims to develop and validate deep learning algorithms capable of classifying skin pigmentation, skin phototype, and photodamage from clinical and dermoscopic skin images. The study will recruit adult participants from multiple countries representing a broad spectrum of skin pigmentation levels, phototypes, and sun exposure patterns. Participants will complete standardized questionnaires, including the Fitzpatrick skin phototype questionnaire and questions related to sun exposure and skin characteristics. Clinical and dermoscopic images will be obtained from predefined anatomical sites on the upper arm and forearm. Skin pigmentation will be assessed using objective measurement methods, including colorimetry and/or spectrophotometry where available, as well as visual classification using the Monk Skin Tone Scale. Trained study personnel will additionally assess Fitzpatrick skin phototype and the degree of photodamage using established clinical scales. Data collected from questionnaires, objective measurements, visual assessments, and imaging will be combined to create reference standards for skin pigmentation, phototype, and photodamage. Deep learning models will subsequently be developed using the collected clinical and dermoscopic images. Model performance will be evaluated against the reference standards using measures of diagnostic accuracy, including sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve. The study will also investigate the agreement between patient-reported assessments, clinician assessments, objective pigmentation measurements, and AI-derived classifications. In addition, external validation studies will be performed using independent dermatological image datasets to assess model robustness and generalizability across different populations, geographic regions, and imaging systems. The anticipated outcome of the study is the development of validated AI algorithms capable of providing standardized and objective assessments of skin pigmentation, phototype, and photodamage. Such tools may support future research, improve characterization of dermatological datasets, facilitate evaluation of AI fairness across skin types, and contribute to more personalized approaches to skin cancer prevention, risk stratification, and dermatological care.

Time frame: At completion of model development and testing (approximately 2029).