Related Experiment Video
Updated: Sep 13, 2025
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
Performance of Artificial Intelligence in Skin Cancer Detection: An Umbrella Review of Systematic Reviews and
Sahand Karimzadhagh1,2,3, Shahriar Ghodous4, Reza M Robati2
1Department of Internal Medicine, School of Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran.
Abstract:
Skin cancer has one of the highest incidence rates among malignancies. A shortage of clinical expertise, particularly in primary care, contrasts with the promising performance of artificial intelligence (AI) models in assisting clinicians. However, a comprehensive evaluation of AI-based diagnostic accuracy across various skin cancers is essential before integration into routine clinical practice. This umbrella review synthesizes evidence from meta-analyses assessing AI model performance in skin cancer detection. We searched PubMed, Web of Science, and Embase for relevant meta-analyses published until January 28, 2025. We included 11 meta-analyses comprising 551 studies from various skin cancer types, clinical settings, and diagnostic modalities. Convolutional neural networks (CNN) and support vector machines (SVM) demonstrated the highest diagnostic performance, with CNN achieving the highest overall accuracy. AI models distinguishing melanoma from melanocytic lesions outperformed those detecting melanoma across all skin cancers, with SVM achieving the highest sensitivity (91%) and specificity (94%). For squamous cell carcinoma, machine learning models trained on hyperspectral imaging demonstrated the highest sensitivity (90.1%) and specificity (92.65%). In differentiating benign from malignant lesions, models exhibited a sensitivity of 87% and a specificity of 86.4%. AI-assisted approaches significantly improved diagnostic accuracy among all clinicians, with generalists and nurse practitioners benefiting more than experienced dermatologists. Deep learning models in primary care, trained on smartphone images, achieved higher sensitivity (90%) and specificity (85%) than general practitioners. AI models significantly outperformed junior dermatologists and nonspecialists compared to senior dermatologists. Hence, integrating AI-assisted tools into clinical workflows, particularly in primary settings, can enhance diagnostic accuracy and minimize missed cases.
