Varga Noémi Nóra, Gulyás Lili, Meznerics Fanni Adél, Barkovskij-Jakobsen Katarina Sofia, Szabó Bence, Hegyi Péter, Bánvölgyi András, Medvecz Márta, Kiss Norbert
Department of Dermatology, Venereology and Dermatooncology, Faculty of Medicine, Semmelweis University, Budapest, Hungary.
Centre for Translational Medicine, Semmelweis University, Budapest, Hungary.
Int J Dermatol. 2025 Oct;64(10):1813-1824. doi: 10.1111/ijd.17828. Epub 2025 May 7.
The incidence of melanoma is increasing worldwide, requiring early detection to improve survival rates. Although dermoscopy is the standard non-invasive tool for diagnosing melanoma, it relies on experience and skill. Advances in optical imaging technologies and artificial intelligence have the potential to improve diagnostic accuracy. Our objective was to compare the diagnostic accuracy of novel non-invasive optical imaging techniques for melanoma detection. A systematic literature search was conducted in three databases (Medline, Embase, and CENTRAL) on November 15, 2023. Inclusion criteria focused on studies comparing the accuracy of optical imaging methods against histopathology. Outcomes consisted of measures of diagnostic accuracy. Random-effects meta-analyses were performed for each method with 95% confidence intervals to summarize all relevant effect sizes. Of the 16,239 records, 141 articles met the inclusion criteria, of which 138 articles were eligible for the meta-analysis. Reflectance confocal microscopy (RCM) and dermoscopy combined with artificial intelligence (DSC + AI) had the highest sensitivity (0.93), with DSC + AI showing higher specificity (0.77 [0.70-0.83]) than RCM (0.749 [0.7475-0.7504]). Multispectral imaging combined with AI also showed high sensitivity (0.92 [0.82-0.97]) and relatively high specificity (0.80 [0.67-0.89]). Standalone dermoscopy exhibited balanced sensitivity (0.87 [0.84-0.90]) and specificity (0.82 [0.78-0.86]). In melanoma diagnosis, both RCM and DSC + AI can serve as second-step optical evaluation methods for suspicious lesions following initial screening with DSC. By maintaining a strong emphasis on multimodal imaging, healthcare providers could improve early detection and outcomes for patients with melanoma.
全球黑色素瘤的发病率正在上升，需要早期检测以提高生存率。尽管皮肤镜检查是诊断黑色素瘤的标准非侵入性工具，但它依赖于经验和技能。光学成像技术和人工智能的进步有可能提高诊断准确性。我们的目标是比较用于黑色素瘤检测的新型非侵入性光学成像技术的诊断准确性。2023年11月15日在三个数据库（Medline、Embase和CENTRAL）中进行了系统的文献检索。纳入标准侧重于比较光学成像方法与组织病理学准确性的研究。结果包括诊断准确性的测量指标。对每种方法进行随机效应荟萃分析，并给出95%置信区间，以总结所有相关效应量。在16239条记录中，141篇文章符合纳入标准（其中138篇文章符合荟萃分析的条件）。反射式共聚焦显微镜（RCM）和皮肤镜检查结合人工智能（DSC+AI）的灵敏度最高（0.93），其中DSC+AI的特异性（0.77[0.70-0.83]）高于RCM（0.749[0.7475-0.7504]）。多光谱成像结合人工智能也显示出高灵敏度（0.92[0.82-0.97]）和相对较高的特异性（0.80[0.67-0.89]）。单独的皮肤镜检查显示出平衡的灵敏度（0.87[0.84-0.90]）和特异性（0.82[0.78-0.86]）。在黑色素瘤诊断中，RCM和DSC+AI均可作为在DSC初始筛查后对可疑病变进行第二步光学评估的方法。通过持续高度重视多模态成像，医疗服务提供者可以改善黑色素瘤患者的早期检测和治疗效果。
