“Pathologist-independent” strategy for T1 colorectal cancer after endoscopic resection
“Pathologist-independent” strategy for T1 colorectal cancer after endoscopic resection
复制标题
T1期结直肠癌内镜切除术后的“独立于病理学家”策略
DOI：
10.1007/s00535-022-01912-5
复制
发表时间：
2022
影响因子：
6.3
通讯作者：
Yeoh Khay Guan
中科院分区：
文献类型：
--
作者：
Ichimasa Katsuro;Kudo Shin-ei;Lee Jonathan Wei Jie;Yeoh Khay Guan
结直肠癌（CRC）筛查计划的广泛采用和内镜治疗的进步，如内镜粘膜下剥离和内镜全层切除术，增加了遇到早期CRC的机会。内镜治疗的粘膜下浸润性（T1）癌症可能需要额外的肠切除术和淋巴结清扫术，这取决于通过以下病理评估确定的淋巴结转移（LNM）风险：粘膜下浸润深度、淋巴血管浸润、分化和肿瘤出芽[1]。目前T1期结直肠癌的临床管理存在两个需要改进的领域[2]：目前预测LNM的诊断能力有限，以及病理学家对四种风险因素的组织学评估诊断不一致。由于先验确定LNM的诊断能力有限，目前的指南提倡对所有具有一个或多个风险因素的T1 CRC进行肠切除和淋巴结清扫，即使LNM的发生率很低（10%）。为了提高T1 CRC中LNM的诊断能力，已经报道了几种基于临床病理信息使用人工智能预测LNM存在的模型，其准确度不同[3]。上述所有四种LNM的危险因素都是病理因素，由病理学家确定。然而，病理学家在评估T1 CRC的淋巴管和血管浸润方面的一致性相当低，其中日本病理学家之间的一致性分别为j= 0.561和j= 0.566，美国病理学家之间的一致性分别为j= 0.518和j= 0.545，欧洲病理学家之间的一致性分别为j= 0.543和j= 0.560 [4]。尽管有报道称使用荧光染色可提高观察者之间在评估淋巴或血管浸润方面的一致性[5]，但目前对荧光染色的使用尚无共识，例如D2-40用于淋巴浸润或维多利亚蓝/Elastica货车Gieson用于血管浸润，因此病理学家之间荧光染色实践的实施和实践存在显著差异。此外，关于粘膜下浸润深度和肿瘤出芽的一致性低于淋巴或血管浸润[6]。这些问题需要更好的解决，如建立一个客观统一的治疗策略。在本期Journal of Gastroenterology中，Song等人开发了一种预测模型，该模型使用深度学习（无手动像素级注释）分析了在9 20倍放大率下扫描的苏木精和伊红（H&E）染色的T1 CRC LNM全切片图像（WSI）[7]。他们的模式成功地克服了上述两个问题。他们的研究包括400例接受内镜切除术和二次手术切除术及淋巴结清扫术的患者，LNM阳性占17.8%（71/400）。主要结果是受试者工作特征曲线下面积
Widespread adoption of colorectal cancer (CRC) screening programs and advances in endoscopic treatments, such as endoscopic submucosal dissection and endoscopic fullthickness resection, have increased the opportunities to encounter early-stage CRC. Endoscopically treated submucosal invasive (T1) cancers may require additional bowel resection with lymph node dissection, depending on the risk of lymph node metastasis (LNM) determined by the following pathologic assessment: depth of submucosal invasion, lymphovascular invasion, differentiation, and tumor budding [1]. The current clinical management of T1 CRC presents two areas for improvement [2]: the limited diagnostic ability currently to predict LNM, and the diagnostic inconsistencies in histologic assessment of the four risk factors among pathologists. Due to the limited diagnostic ability to a priori determine LNM, current guidelines advocate intestinal resection with lymph node dissection for all T1 CRC with one or more risk factors, even despite the low rates of LNM (10%). To better the diagnostic ability of LNM among T1 CRC, several models predicting the presence of LNM using artificial intelligence based on clinicopathological information have been reported with varying degrees of accuracy [3]. All four of the aforementioned risk factors for LNM are pathological factors and are determined by pathologists. However, there has been fairly low agreement among pathologists in assessing lymphatic and vascular invasion in T1 CRCs, whereby the agreement among Japanese pathologists was j= 0.561 and j= 0.566, and j= 0.518 and j= 0.545 among American pathologists, and j= 0.543 and j= 0.560 among Europe pathologists, respectively [4]. Although the use of immunohistostaining has been reported to increase interobserver agreement in evaluating the presence of lymphatic or vascular invasion [5], there is currently no consensus for the use of immunohistostaining, such as D2-40 for lymphatic invasion or Victoria Blue/Elastica van Gieson for vascular invasion, whereby the conduct and practice of immunohistostaining practice vary significantly among pathologists. Furthermore, there was less agreement regarding the depth of submucosal invasion and tumor budding than for lymphatic or vascular invasion [6]. These issues need better solutions, such as to establish an objective and uniform treatment strategy. In this issue of Journal of Gastroenterology, Song et al. developed a prediction model analyzing hematoxylin and eosin (H&E)-stained whole slide images (WSIs) scanned at 9 20 magnification for LNM in T1 CRC using deeplearning without manual-pixel-level annotation [7]. Their model succeeded in overcoming the aforementioned two problems. Their study included 400 patients who underwent endoscopic resection with secondary surgical resection with lymph node dissection, and LNM-positivity accounted for 17.8%(71/400). The main outcome was the area under the receiver operating characteristic curve