AI-Derived Facial Redness Score and Body Mass Index in a Large Japanese Screening Cohort
Description
This dataset supports the research letter "AI-Derived Facial Redness Score and Body Mass Index in a Large Japanese Screening Cohort." It comprises fully anonymized, non-identifiable data from adults who participated in the "Karada Measurement" public health screening program at the Osaka Healthcare Pavilion, Expo 2025 Osaka, Kansai, Japan. Of 484,462 individuals assessed for eligibility, 74,078 were excluded owing to missing body mass index (BMI) data (n=50,293) or missing facial redness scores (n=23,785), yielding a final analytical cohort of 410,384 participants (250,318 women and 160,066 men; mean age, 44.4 years; mean BMI, 22.3 kg/m²). Facial skin characteristics were quantified using an AI-based skin analysis system (Perfect Corp., Taiwan), which generated relative scores (range, 1–100) for 14 facial parameters: spots, wrinkles, texture, oiliness, redness, eye bags, pores, moisture, upper eyelid drooping, lower eyelid drooping, firmness, radiance, dark circles, and acne. The redness score represents an image-derived facial redness measure (not dermatologist-assessed erythema) and was reverse-coded (reverse-coded score = 101 − original score) so that higher values denote greater facial redness. Vascular age and pulse rate were additionally measured using a contact-type infrared photoplethysmographic sensor. BMI was categorized according to Japanese criteria. The data were used to examine the association between BMI and AI-derived facial redness via Pearson correlations and age-adjusted logistic regression, including sensitivity analyses across five redness thresholds and exploratory subgroup analyses (women aged 30–59 years and age-matched men). This study was approved by the Institutional Review Board of Kindai University Hospital (Approval No. 2025-220) and conducted in accordance with the Declaration of Helsinki. All participants provided informed consent for the research use of their data at the time of collection. The dataset is fully anonymized and contains no identifiable personal information or facial images.