人工智能在系统性硬化症中的应用:临床应用、挑战与未来方向
本叙述性综述系统审视了人工智能在系统性硬化症(SSc)中的当前应用,涵盖机器学习、深度学习和影像组学在疾病表型分型、风险分层及影像量化方面的潜力,重点讨论了AI驱动的高分辨率CT分析、甲襞毛细血管镜计算机视觉评估及皮肤病变的AI辅助量化,并指出多数应用仍处于探索阶段,受限于小样本、单中心数据和外部验证不足。
要点
- AI在SSc-ILD的HRCT纤维化自动分割和影像组学风险建模中展现潜力
- 计算机视觉用于甲襞毛细血管镜可实现微血管分期的专家级一致性
- 多数AI应用仍为探索性,受限于小数据、数据异质性和外部验证不足
核心结论
- AI在SSc的表型分型、风险分层和影像量化方面展现潜力,但多数应用仍为探索性且受限于稀有病的「小数据」约束,尚不能立即临床转化。
研究设计与样本
- 本文为叙述性综述,批判性审视AI在系统性硬化症中的当前应用、方法学局限、转化就绪度和未来方向。
主要发现
- AI驱动的高分辨率CT分析可实现SSc-ILD的自动纤维化分割和影像组学风险建模。
- 计算机视觉应用于甲襞毛细血管镜在微血管分期上达到专家级一致性,新兴数字工具旨在量化雷诺现象动态。
- 在皮肤病变方面,AI辅助超声、光学成像和组织病理学可提供真皮重塑的可重复量化。
- 多数应用仍为探索性,基于小型或单中心数据集,外部验证有限。
- 数据异质性、标注负担和稀有病的「小数据」约束限制了即时临床转化。
临床怎么用
- AI工具目前不宜替代标准临床评估,应在研究中用于辅助表型分型和影像量化探索。
- 临床引入AI前需关注多中心验证、数据异质性和伦理治理框架。
局限性
- 多数AI应用基于小型或单中心数据集,外部验证有限。
- 数据异质性、标注负担和稀有病的「小数据」约束限制临床转化。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Systemic sclerosis (SSc) is a rare autoimmune disease defined by immune dysregulation, vasculopathy, and progressive fibrosis of the skin and internal organs. Despite advances in care, major complications such as interstitial lung disease (ILD) and myocardial involvement remain the leading causes of morbidity and mortality. Current assessment tools, such as the modified Rodnan skin score, pulmonary function tests, and nailfold capillaroscopy, are limited by subjectivity and interobserver variability. Artificial intelligence (AI) is reshaping this landscape. Machine learning, deep learning, and radiomics have shown potential to enhance disease phenotyping, risk stratification, and imaging quantification in SSc. AI-driven high-resolution computed tomography analysis enables automated fibrosis segmentation and radiomic risk modeling in SSc-ILD. Computer vision approaches applied to nailfold capillaroscopy achieve expert-level agreement for microvascular staging, whereas emerging digital tools aim to quantify Raynaud phenomenon dynamics. In skin disease, AI-assisted ultrasonography, optical imaging, and histopathology provide reproducible quantification of dermal remodeling. However, most applications remain exploratory, based on small or single-center data sets with limited external validation. Challenges such as data heterogeneity, annotation burden, and the "small data" constraints of rare diseases limit immediate clinical translation. This narrative review critically examines current AI applications in SSc, highlighting methodologic limitations, translational readiness, and future directions toward robust, multicenter, and ethically governed implementation.
引用
- Sieiro Santos C, Sequí-Sabater JM, Benavent D, et al. Artificial Intelligence in Systemic Sclerosis: Clinical Applications, Challenges, and Future Directions. Arthritis Care Res (Hoboken). 2026. PMID: 42007790. doi:10.1002/acr.80068
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。