AI皮肤癌检测存在肤色偏差，需改进评估方法以实现公平
Towards Fairness in AI for Melanoma Detection: Systemic Review and Recommendations
- 用洛蕾尔色卡图结合肤色色调，更全面评估皮肤色度
- 2013–2024年研究中，轻肤色数据占比超75%，模型对深肤色效果差
- 提出医疗AI公平性框架，指导未来研究设计与评估
早期准确的黑色素瘤检测对改善患者预后至关重要。近年来人工智能在该领域展现出潜力，但其在不同肤色人群中的有效性仍面临严峻挑战。本研究系统回顾了2013至2024年间基于深度学习的皮肤癌检测研究，聚焦于算法方法、数据集及肤色表征。结果显示，尽管AI可提升检测效率，但模型普遍偏向浅肤色人群。为此，我们建议在现有洛蕾尔色卡图（L'Oréal Color Chart Map）基础上引入肤色色调（skin hue），以实现更全面的肤色评估。研究强调，必须采用多样化数据集和稳健评估指标，推动开发对所有患者均公平有效的AI模型。通过采纳专为医疗与黑色素瘤检测设计的PRISMA Equity框架，有望缩小黑色素瘤诊疗中的不平等现象。
原文摘要 · Abstract (English)
Early and accurate melanoma detection is crucial for improving patient outcomes. Recent advancements in artificial intelligence AI have shown promise in this area, but the technologys effectiveness across diverse skin tones remains a critical challenge. This study conducts a systematic review and preliminary analysis of AI based melanoma detection research published between 2013 and 2024, focusing on deep learning methodologies, datasets, and skin tone representation. Our findings indicate that while AI can enhance melanoma detection, there is a significant bias towards lighter skin tones. To address this, we propose including skin hue in addition to skin tone as represented by the LOreal Color Chart Map for a more comprehensive skin tone assessment technique. This research highlights the need for diverse datasets and robust evaluation metrics to develop AI models that are equitable and effective for all patients. By adopting best practices outlined in a PRISMA Equity framework tailored for healthcare and melanoma detection, we can work towards reducing disparities in melanoma outcomes.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。