Artificial intelligence in dermatology: a comparative analysis of computer vision programs based on machine learning models

Abstract

Objective: To compare modern computer programs (smartphone programs – mobile applications) using artificial intelligence (AI) for diagnosing and dynamic monitoring of skin conditions. Material and methods. A total of 1,319 publications were identified for AI-powered computer programs using targeted searches in PubMed/MEDLINE and Google Scholar databases, as well as in the eLibrary and CyberLeninka electronic libraries for the period 2016–2025. Queries focused on AI, convolutional neural networks (CNNs), computer programs (mobile apps), and dermatovenereology were used. After a multi-stage screening based on inclusion/exclusion criteria (including the availability of quantitative performance metrics), 9 key articles with specific descriptions of the computer programs (mobile apps) were selected. A search and subsequent analysis identified 9 computer programs (Google DermAssist, SkinIO, Melanoma Check, Derma Onko Check, SkinVision, Tibot, SkinScan, Aysa, and Skinive), which use AI to diagnose and monitor skin conditions. Results. Effectiveness of the programs varies: Google DermAssist and Derma Onko Check demonstrated high accuracy (96–97%) and sensitivity (97–98%), while Skinive showed improvement in metrics over time from 2020 to 2021 (maximum sensitivity of 97.9% and specificity of 97.1%). Limitations include dependence on photo image quality, low effectiveness for rare conditions and dark skin tones, and the need for a biopsy to confirm a diagnosis. Mobile apps using CNN demonstrate high sensitivity (87–97.9%), though specificity varies significantly (70–98%), which may increase the number of additional consultations with specialist doctors when using these programs in diagnostics. Conclusion . AI-based software (mobile apps) offers significant potential for increasing the accessibility and accuracy of skin pathology diagnostics, especially in remote areas and regions with a shortage of dermatovenereologists. Promising developments encompass the integration of computer programs with telemedicine, the refinement of algorithms for diagnosing rare pathologies, and the standardization of testing to enhance result reproducibility.

Ähnliche Arbeiten