色素性病变数字皮肤镜监测:两种结合静态与动态特征识别黑色素瘤的新计算工具
本研究收集数字皮肤镜监测后切除的173处色素性病变(65处黑色素瘤、108处痣)的皮肤镜图像,进行静态与动态特征评估,基于逻辑回归和随机森林构建两种多变量预测模型,AUC分别为0.87和0.89,高于静态评估模型(ABCD TDS评分0.57、7点检查表0.59),并转化为两种风险计算器。
要点
- 逻辑回归与随机森林模型的AUC分别为0.87和0.89,高于ABCD TDS评分0.57和7点检查表0.59。
- 研究纳入173处病变,其中黑色素瘤65处、痣108处,42处黑色素瘤为原位。
- 作者提出两种诊断工具,称可显著提高数字皮肤镜监测中鉴别黑色素瘤与痣的准确性。
核心结论
- 整合色素性病变的动态与静态评估是一种安全的方法,可显著提高皮肤黑色素瘤的诊断准确性。
- 作者提出两种诊断工具,在数字皮肤镜监测中可显著提高鉴别黑色素瘤与痣的准确性。
研究设计与样本
- 收集数字皮肤镜监测后切除的色素性病变的皮肤镜图像,并对皮肤镜特征进行静态和动态评估。
- 基于逻辑回归和随机森林构建两种多变量预测模型。
- 共评估173处病变,包括65处皮肤黑色素瘤和108处痣。
主要发现
- 42处黑色素瘤为原位,浸润性黑色素瘤的中位厚度为0.35 mm。
- 黑色素瘤的中位随访时间为9.8个月,痣为9.1个月。
- 逻辑回归模型和随机森林模型的AUC分别为0.87和0.89, substantially高于静态评估模型(ABCD TDS评分0.57;7点检查表0.59)。
关键数值与推荐
- 逻辑回归模型AUC为0.87,随机森林模型AUC为0.89。
- ABCD TDS评分AUC为0.57,7点检查表AUC为0.59。
- 黑色素瘤中位随访时间9.8个月,痣中位随访时间9.1个月。
- 浸润性黑色素瘤中位厚度0.35 mm。
临床怎么用
- 作者构建了两种风险计算器,将所提出的模型转化为用户友好型应用,以辅助临床医生的决策过程。
- 作者认为整合动态与静态评估是提高皮肤黑色素瘤诊断准确性的安全方法。
局限性
- 原文未报告局限性。
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
BACKGROUND: Early diagnosis is the most effective intervention to improve the prognosis of cutaneous melanoma. Even though the introduction of dermoscopy has improved the diagnostic accuracy, it can still be difficult to distinguish some melanomas from benign melanocytic lesions. Digital dermoscopy monitoring can identify dynamic changes of melanocytic lesions: To date, some algorithms were proposed, but a universally accepted one is still lacking. OBJECTIVES: To identify independent predictive variables associated with the diagnosis of cutaneous melanoma and develop a multivariable dermoscopic prediction model able to discriminate benign from malignant melanocytic lesions undergoing digital dermoscopy monitoring. METHODS: We collected dermoscopic images of melanocytic lesions excised after dermoscopy monitoring and carried out static and dynamic evaluations of dermoscopic features. We built two multivariable predictive models based on logistic regression and random forest. RESULTS: We evaluated 173 lesions (65 cutaneous melanomas and 108 nevi). Forty-two melanomas were in situ, and the median thickness of invasive melanomas was 0.35 mm. The median follow-up time was 9.8 months for melanomas and 9.1 for nevi. The logistic regression and random forest models performed with AUC values of 0.87 and 0.89, respectively, were substantially higher than those of the static evaluation models (ABCD TDS score, 0.57; 7-point checklist, 0.59). Finally, we built two risk calculators, which translate the proposed models into user-friendly applications, to assist clinicians in the decision-making process. CONCLUSIONS: The present study demonstrates that the integration of dynamic and static evaluations of melanocytic lesions is a safe approach that can significantly boost the diagnostic accuracy for cutaneous melanoma. We propose two diagnostic tools that significantly increase the accuracy in discriminating melanoma from nevi during digital dermoscopy monitoring.
引用
- Zenone M, Zocchi L, Moccia C, et al. Digital dermoscopy monitoring of melanocytic lesions: Two novel calculators combining static and dynamic features to identify melanoma. J Eur Acad Dermatol Venereol. 2021. PMID: 34862986. doi:10.1111/jdv.17852
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。