文献速递骨软组织肉瘤与黑色素瘤

传统图像分析器与卷积神经网络在皮肤镜诊断中的前瞻性头对头比较

Past and present of computer-assisted dermoscopic diagnosis: performance of a conventional image analyser versus a convolutional neural network in a prospective data set of 1,981 skin lesions.
PubMed 文献 · Eur J Cancer · 2020-06-10

本研究基于1981处皮肤病变的前瞻性皮肤镜数据集,将市场批准的卷积神经网络(CNN,Moleanalyzer-Pro™)与依赖手工特征分割加权的传统图像分析器(CIA,Moleanalyzer-3™/Dynamole™)进行头对头比较。CNN的敏感度77.6%、特异度95.3%、ROC-AUC 0.945,均显著优于CIA的53.4%、86.6%和0.738(均p<0.001)。作者认为CNN诊断性能更优,不支持继续将传统CIA作为临床决策辅助工具。

要点

  • CNN敏感度77.6%、特异度95.3%、AUC 0.945,均显著优于传统图像分析器。
  • 传统CIA敏感度仅53.4%,特异度86.6%,AUC 0.738。
  • 结论不支持继续将传统图像分析器用于辅助临床决策。

核心结论

  • CNN在皮肤镜病变分类中的诊断性能显著优于依赖手工特征的传统图像分析器,提示不应继续将后者作为医生临床管理决策的辅助工具。

研究设计与样本

  • 横断面研究,使用真实世界前瞻性采集的1981处皮肤病变皮肤镜数据集,对市场批准的CNN与传统CIA进行头对头比较。
  • 参考标准为组织病理诊断(n=785),未切除的良性病变(n=1196)则采用专家共识加至少2年序贯数字皮肤镜随访无异常事件。
  • 数据集来自435例患者(62.2%男性,平均年龄52岁),含281处恶性病变与1700处良性病变。

主要发现

  • CNN敏感度77.6%(95%CI 72.4%-82.1%)、特异度95.3%(95%CI 94.2%-96.2%)、ROC-AUC 0.945(95%CI 0.930-0.961)。
  • CIA敏感度53.4%(95%CI 47.5%-59.1%)、特异度86.6%(95%CI 84.9%-88.1%)、ROC-AUC 0.738(95%CI 0.701-0.774)。
  • 敏感度、特异度与ROC-AUC的两两比较均显示CNN明显更优(均p<0.001)。
  • 数据集中包含经序贯数字皮肤镜动态变化初诊的黑色素瘤(201例中52例,20.6%),该部分降低了两个分类器的敏感度。

临床怎么用

  • 在皮肤镜辅助诊断场景中,应优先选择基于深度学习的CNN类工具,而非依赖手工特征分割与加权的传统图像分析器。
  • 对经序贯数字皮肤镜动态变化发现的黑色素瘤需保持警惕,此类病变可降低分类器敏感度。

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: Convolutional neural networks (CNNs) have shown a dermatologist-level performance in the classification of skin lesions. We aimed to deliver a head-to-head comparison of a conventional image analyser (CIA), which depends on segmentation and weighting of handcrafted features, to a CNN trained by deep learning. METHODS: Cross-sectional study using a real-world, prospectively acquired, dermoscopic dataset of 1981 skin lesions to compare the diagnostic performance of a market-approved CNN (Moleanalyzer-Pro™, developed in 2018) to a CIA (Moleanalyzer-3™/Dynamole™; developed in 2004, all FotoFinder Systems Inc, Germany). As a reference standard, we used histopathological diagnoses (n = 785) or, in non-excised benign lesions (n = 1196), expert consensus plus an uneventful follow-up by sequential digital dermoscopy for at least 2 years. RESULTS: A total of 281 malignant lesions and 1700 benign lesions from 435 patients (62.2% male, mean age: 52 years) were prospectively imaged. The CNN showed a sensitivity of 77.6% (95% confidence interval [CI]: [72.4%-82.1%]), specificity of 95.3% (95% CI: [94.2%-96.2%]), and receiver operating characteristic (ROC)-area under the curve (AUC) of 0.945 (95% CI: [0.930-0.961]). In contrast, the CIA achieved a sensitivity of 53.4% (95% CI: [47.5%-59.1%]), specificity of 86.6% (95% CI: [84.9%-88.1%]) and ROC-AUC of 0.738 (95% CI: [0.701-0.774]). The data set included melanomas originally diagnosed by dynamic changes during sequential digital dermoscopy (52 of 201, 20.6%), which reduced the sensitivities of both classifiers. Pairwise comparisons of sensitivities, specificities, and ROC-AUCs indicated a clear outperformance by the CNN (all p < 0.001). CONCLUSIONS: The superior diagnostic performance of the CNN argues against a continued application of former CIAs as an aide to physicians' clinical management decisions.

来源:Eur J Cancer,查看原文。版权归原作者所有。

引用

  1. Sies K, Winkler JK, Fink C, et al. Past and present of computer-assisted dermoscopic diagnosis: performance of a conventional image analyser versus a convolutional neural network in a prospective data set of 1,981 skin lesions. Eur J Cancer. 2020. PMID: 32534243. doi:10.1016/j.ejca.2020.04.043

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。