2nd Edition of Dermatology and Plastic Surgery World Conference 2026

Abstract

Dermatological AI models exhibit critical racial bias 95.5percent of training data derives from Fitzpatrick I to III populations, causing a documented 15-25% drop in diagnostic sensitivity for melanin rich skin (Fitzpatrick IVVI). A compounding modality gap existsGPT4 scores 90percent on text vignettes but only 57percent on clinical images, with this 33point disparity widening further for pigmented lesions. Simultaneously, 72percent of viral skin remedies on social media are harmful, causing chemical burns, hyperpigmentation, and Koebnerization disproportionately in dark skinned populations. TVMCH OPD receives over 200 dermatology cases per day with over 99 percent of patients presenting Fitzpatrick IVVI skin, making it an ideal validation hub for decolonised AI benchmarking.

VARNAM is a crosssectional comparative benchmarking study designed to benchmark the top1 diagnostic accuracy of GPT4o, Gemini 2.0, and Claude 3.5 Sonnet across five South Indian pigmentary disorders (Melasma, PostAcne PIH, Vitiligo, Acanthosis Nigricans, Lichen Planus Pigmentosus), quantify the Dangerous Recommendation Rate (DRR) against IADVL consensus guidelines, and audit AI capability to proactively debunk regional social media health myths (M3 Axis). Group A (n is equal to100) comprises TVMCH confirmed cases, Fitzpatrick IVVI, with 20 cases per diagnosis Group B (n is equal to100) comprises DermNet NZ or HAM10000 phenotypically matched Fitzpatrick I to III controls. Deidentified clinical images and structured vignettes will be queried 3 per case via official APIs at temperature 0 with a 14-day collection window and 2to3 majority vote rule. Gold standard is defined as complete diagnostic consensus by 3 Senior

Dermatology Consultants or archived histopathology  inter rater reliability target is Fleiss Kappa greater than  0.81. Full IEC clearance from TVMCH Ethics Committee has been obtained the study is compliant with ICMR 2017 National Ethical Guidelines and DPDP Act 2023, with mandatory de identification of all patient images before API upload. VARNAM provides the first standardised tri modal benchmarking framework for South Indian pigmentary disorders, generating directly actionable evidence for equitable AI deployment in dark skin predominant healthcare settings across India and the broader Global South.