Deep learning model for the prediction of microsatellite instability in colorectal cancer: a diagnostic study
Deep learning model for the prediction of microsatellite instability in colorectal cancer: a diagnostic study
复制标题
DOI：
10.1016/s1470-2045(20)30535-0
复制
发表时间：
2021-01-01
期刊：
影响因子：
51.1
通讯作者：
Shen, Jeanne
中科院分区：
文献类型：
--
作者：
Yamashita, Rikiya;Long, Jin;Shen, Jeanne
背景检测结直肠癌微卫星不稳定性(MSI)对临床决策至关重要，因为它可以识别不同治疗反应和预后的患者。建议进行通用的MSI检测，但许多患者仍未进行检测。存在着对广泛可用、成本效益高的工具的迫切需求，以帮助患者选择检测。在这里，我们研究了基于深度学习的系统直接从苏木精伊红(H&E)染色的全幻灯片图像(WSIS)中自动预测MSI的潜力。方法我们的深度学习模型(MSINet)使用100个H&E染色的WSIS(50个具有微卫星稳定性[MSS]和50个具有MSI)，以40倍的放大倍数扫描，每个患者以分类平衡的方式从斯坦福大学医学中心(美国加州斯坦福；内部数据集)接受初次结直肠癌切除的343名患者中随机选择一名患者。我们在抵抗测试集(来自15名患者的15个H&E染色的WSIS；7个MSS和8个MSI)上内部验证了该模型，并在外部验证了来自癌症基因组图谱的484个H&E染色的WSIS(402例MSS和77个MSI；479名患者)，其中包含以40倍和20倍放大扫描的WSIS。主要使用灵敏度、特异度、阴性预测值(NPV)和受试者工作特征曲线下面积(AUROC)来评估性能。我们将该模型与5位胃肠道病理学家在40倍放大倍率的外部数据集中随机选择的子集上的性能进行了比较，发现MSINet模型在内部数据集的抵抗测试集上的AUROC为0.931(95%CI 0.771-1-000)，在外部数据集上的AUROC为0.779(0.720-0.838)。在外部数据集上，使用灵敏度加权操作点，模型获得了93.7%(95%CI 90.3-96.2)的NPV，76.0%(64.8-85.1)的敏感度和66.6%(61.8-71.2)的特异度。在阅读器实验(40例)上，该模型获得了0.865的AUROC(95%CI 0.735-0.995)。五位病理学家的平均AUROC表现为0.605(95%CI 0.453-0.757)。解释我们的深度学习模型在预测H&E染色的WSI上的MSI方面超过了经验丰富的胃肠道病理学家的表现。在目前通用的MSI测试范例中，这种模型可能会作为一种自动筛查工具对患者进行验证性测试，潜在地减少测试患者的数量，从而节省大量与测试相关的劳动力和成本。版权所有(C)2020爱思唯尔有限公司。保留所有权利。
Background Detecting microsatellite instability (MSI) in colorectal cancer is crucial for dinical decision making, as it identifies patients with differential treatment response and prognosis. Universal MSI testing is recommended, but many patients remain untested. A critical need exists for broadly accessible, cost-efficient tools to aid patient selection for testing. Here, we investigate the potential of a deep learning-based system for automated MSI prediction directly from haematoxylin and eosin (H&E)-stained whole-slide images (WSIs).Methods Our deep learning model (MSINet) was developed using 100 H&E-stained WSIs (50 with microsatellite stability [MSS] and 50 with MSI) scanned at 40x magnification, each from a patient randomly selected in a classbalanced manner from the pool of 343 patients who underwent primary colorectal cancer resection at Stanford University Medical Center (Stanford, CA, USA; internal dataset) between Jan 1,2015, and Dec 31,2017. We internally validated the model on a holdout test set (15 H&E-stained WSIs from 15 patients; seven cases with MSS and eight with MSI) and externally validated the model on 484 H&E-stained WSIs (402 cases with MSS and 77 with MSI; 479 patients) from The Cancer Genome Atlas, containing WSIs scanned at 40x and 20x magnification. Performance was primarily evaluated using the sensitivity, specificity, negative predictive value (NPV), and area under the receiver operating characteristic curve (AUROC). We compared the model's performance with that of five gastrointestinal pathologists on a class-balanced, randomly selected subset of 40x magnification WSIs from the external dataset (20 with MSS and 20 with MSI).Findings The MSINet model achieved an AUROC of 0.931 (95% CI 0.771-1-000) on the holdout test set from the internal dataset and 0.779 (0.720-0 .838) on the external data set. On the external dataset, using a sensitivity-weighted operating point, the model achieved an NPV of 93.7% (95% CI 90.3-96.2), sensitivity of 76.0% (64.8-85.1), and specificity of 66.6% (61.8-71.2). On the reader experiment (40 cases), the model achieved an AUROC of 0.865 (95% CI 0.735-0.995). The mean AUROC performance of the five pathologists was 0.605 (95% CI 0.453-0.757).Interpretation Our deep learning model exceeded the performance of experienced gastrointestinal pathologists at predicting MSI on H&E-stained WSIs. Within the current universal MSI testing paradigm, such a model might contribute value as an automated screening tool to triage patients for confirmatory testing, potentially reducing the number of tested patients, thereby resulting in substantial test-related labour and cost savings. Copyright (C) 2020 Elsevier Ltd. All rights reserved.