Artificial intelligence-powered teledermatology solution for skin cancer screening using 3D total body photography images

Summary: Abstract Body: Skin cancer, including Basal Cell Carcinoma (BCC), squamous cell carcinoma (SCC), and melanoma, poses a significant health challenge, with early detection being critical for better outcomes. This study utilizes Artificial Intelligence (AI) to differentiate between benign and malignant lesions using 3D total-body photography (3D-TBP) images, whose quality closely resembles that of smartphone images commonly used in telemedicine. By enabling effective skin cancer screening in non-specialist settings, this approach aims to expand teledermatology initiatives, particularly in underserved regions. The SLICE-3D dataset, by the International Skin Imaging Collaboration, was used for model development. This dataset comprises 629,941 images, which includes 302,771 benign and 327,170 malignant acquired via 3D-TBP systems and includes comprehensive lesion phenotypes from diverse populations across six continents. The data were split into an 80:20 training-to-testing ratio, and AI models—including Random Forest, Logistic Regression, ResNet, Support Vector Machines, and Convolutional Neural Networks (CNN)—were evaluated. CNNs achieved the best performance with 99.3% accuracy, 80% sensitivity, 99.4% specificity, and F1 score of 0.99. By utilizing the "ugly duckling sign," which identifies outlier lesions against a patient's baseline phenotype, the AI models address lesion-selection bias inherent in traditional dermoscopic datasets. This approach ensures accurate differentiation of malignant lesions, even with lower-resolution images. Moreover, its adaptability to smartphone-quality photos enhances its potential for broader telemedicine applications. This study demonstrates the transformative role of AI-powered teledermatology in expanding early skin cancer detection in underserved regions. By enabling timely triage and intervention, this approach could significantly reduce the global burden of skin cancer. Sakthi Jaya Sundar Rajasekar¹, Hemanth N², Varalakshmi Perumal³ 1. Melmaruvathur Adhiparasakthi Institute of Medical Sciences and Research, Melmaruvathur, TN, India. 2. Madras Institute of Technology, Chennai, TN, India. 3. Anna University College of Engineering Guindy, Guindy, TN, India. Bioinformatics, Computational Biology, and Imaging