基于卷积神经网络的内镜图像分析诊断胃癌及预测浸润深度的诊断准确性:系统评价与Meta分析
该系统评价与Meta分析纳入17项研究(51,446张图像、174段视频、5539例患者),评估卷积神经网络(CNN)诊断胃癌及预测浸润深度的准确性,并比较其与内镜医师的表现。CNN诊断胃癌的汇总敏感度89%、特异度93%、AUC 0.94;预测浸润深度的汇总敏感度82%、特异度90%、AUC 0.90。CNN诊断胃癌的表现与专家内镜医师无显著差异(0.95 vs 0.90,P > .05),优于整体内镜医师(0.95 vs 0.87,P < .05)。
要点
- CNN诊断胃癌汇总敏感度89%、特异度93%、AUC 0.94。
- CNN预测胃癌浸润深度汇总敏感度82%、特异度90%、AUC 0.90。
- CNN诊断胃癌表现与专家内镜医师无显著差异(0.95 vs 0.90,P > .05),优于整体内镜医师(0.95 vs 0.87,P < .05)。
核心结论
- CNN诊断胃癌及预测胃癌浸润深度具有较高准确性,诊断胃癌的表现与专家内镜医师无显著差异。
- 需要开展CNN诊断胃癌实时性能的研究以证实上述发现。
研究设计与样本
- 系统评价与Meta分析,检索PubMed、Embase、Web of Science及灰色文献至2021年7月23日。
- 纳入17项研究,共51,446张图像、174段视频、5539例患者。
主要发现
- CNN诊断胃癌的汇总敏感度89%(95%CI 85-93)、特异度93%(95%CI 88-97)、阳性似然比13.4(95%CI 7.3-25.5)、阴性似然比0.11(95%CI 0.07-0.17)、AUC 0.94(95%CI 0.91-0.98)。
- CNN预测胃癌浸润深度的汇总敏感度82%(95%CI 78-85)、特异度90%(95%CI 82-95)、阳性似然比8.4(95%CI 4.2-16.8)、阴性似然比0.20(95%CI 0.16-0.26)、AUC 0.90(95%CI 0.87-0.93)。
- CNN诊断胃癌的AUC与专家内镜医师无显著差异(0.95 vs 0.90,P > .05),高于整体内镜医师(0.95 vs 0.87,P < .05)。
局限性
- 需要开展CNN诊断胃癌实时性能的研究以证实上述发现。
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 AND AIMS: This study aimed to evaluate the accuracy and effectiveness of the convolutional neural network (CNN) in diagnosing gastric cancer and predicting the invasion depth of gastric cancer and to compare the performance of the CNN with that of endoscopists. METHODS: PubMed, Embase, Web of Science, and gray literature were searched until July 23, 2021 for studies that assessed the diagnostic accuracy of CNN-assisted examinations for gastric cancer or the invasion depth of gastric cancer. Studies meeting inclusion criteria were included in the systematic review and meta-analysis. RESULTS: Seventeen studies comprising 51,446 images and 174 videos of 5539 patients were included. The pooled sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR-), and area under the curve (AUC) of the CNN for diagnosing gastric cancer were 89% (95% confidence interval [CI], 85-93), 93% (95% CI, 88-97), 13.4 (95% CI, 7.3-25.5), .11 (95% CI, .07-.17), and .94 (95% CI, .91-.98), respectively. The performance of the CNN in diagnosing gastric cancer was not significantly different from that of expert endoscopists (.95 vs .90, P > .05) and was better than that of overall endoscopists (experts and nonexperts) (.95 vs .87, P < .05). The pooled sensitivity, specificity, LR+, LR-, and AUC of the CNN for predicting the invasion depth of gastric cancer were 82% (95% CI, 78-85), 90% (95% CI, 82-95), 8.4 (95% CI, 4.2-16.8), .20 (95% CI, .16-.26), and .90 (95% CI, .87-.93), respectively. CONCLUSIONS: The CNN is highly accurate in diagnosing gastric cancer and predicting the invasion depth of gastric cancer. The performance of the CNN in diagnosing gastric cancer is not significantly different from that of expert endoscopists. Studies of the real-time performance of the CNN for gastric cancer diagnosis are needed to confirm these findings.
引用
- Xie F, Zhang K, Li F, et al. Diagnostic accuracy of convolutional neural network-based endoscopic image analysis in diagnosing gastric cancer and predicting its invasion depth: a systematic review and meta-analysis. Gastrointest Endosc. 2021. PMID: 34979114. doi:10.1016/j.gie.2021.12.021
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。