EP02.46: An AI system (SonoLyst) achieves expert level performance when categorising images for adherence to ISUOG mid-trimester screening guidelines
EP02.46: An AI system (SonoLyst) achieves expert level performance when categorising images for adherence to ISUOG mid-trimester screening guidelines
复制标题
EP02.46：人工智能系统 (SonoLyst) 在对图像进行分类以遵守 ISUOG 中期筛查指南时实现了专家级性能
DOI：
10.1002/uog.26661
复制
发表时间：
2023
影响因子：
7.1
通讯作者：
Papageorghiou A
中科院分区：
文献类型：
--
作者：
Papageorghiou A
目的 评估基于人工智能 (AI) 的算法 SonoLyst（GEHC、Zipf）在根据 ISUOG 指南对完整常规孕中期超声扫描所需图像进行“同行评审”方面的性能。方法使用在常规异常扫描期间从 4 个站点捕获的真实世界图像进行多中心前瞻性评估。这些内容由从之前的视频循环中提取的非最佳视图进行补充，以允许更可靠的测试。为了复制常规临床图像审查，每张图像都被分类到相关的 ISUOG 视图中，然后评估协议遵守情况的质量。每张图像均由人工智能和 12 名临床专家中的 4 名临床专家进行评估，对彼此的结果不知情。将专家与专家之间的协议与专家与 SonoLyst 之间的协议进行了比较。根据操作员经验进行子分析。结果我们评估了来自 776 名患者的 12,476 张图像。专家与专家对视图进行分类的平均一致性为 79.4%（77.6% 至 81.1%），与专家与 SonoLyst 的一致性为 79.6%（77.1% 至 82.4%）没有区别。 结论 在根据 ISUOG 指南对常规孕中期扫描的视图进行分类时，SonoLyst 与专家没有区别。此类软件可以在常规扫描期间支持超声检查人员，通过捕获图像的“实时同行评审”来确保扫描的完整性和质量。
ObjectivesTo evaluate the performance of SonoLyst (GEHC, Zipf), an Artificial Intelligence (AI) based algorithm, in undertaking “peer review” of images required for complete routine mid-trimester ultrasound scans according to ISUOG guidelines.MethodsMulticentre prospective evaluation using real-world images captured from 4 sites during routine anomaly scanning. These were supplemented by non-optimal views extracted from preceding video loops to allow more robust testing. To replicate routine clinical image review, each image was categorised into the relevant ISUOG view, and then quality assessed for protocol adherence. Each image was assessed by AI and 4 clinical experts from a pool of 12, blinded to each other's results. Expert-expert agreement was compared to expert-SonoLyst agreement. Subanalysis by operator experience was undertaken.ResultsWe assessed 12,476 images from 776 patients. Mean Expert-Expert agreement in categorising views was 79.4%(77.6% to 81.1%) and was indistinguishable to Expert-SonoLyst agreement of 79.6%(77.1% to 82.4%).ConclusionsSonoLyst is indistinguishable from an expert when categorising views of routine mid-trimester scans against ISUOG guidelines. Such software can support sonographers during routine scans to ensure completeness and quality of scanning by “real-time peer review” of captured images.