编者按：2017年12月12日，《美国医学会杂志》连发四篇有关人工智能的文章，其中三篇与乳腺癌相关，尤其第一篇的研究者挑战赛令人叹为观止，共有390支来自世界各地的参赛队报名，最后哈佛医学院、麻省理工学院、麻省总医院、香港中文大学、德国慕尼黑ExB胜出！
机器学习是人工智能的分支，即以推理为主→以知识为主→以学习为主实现人工智能的途径。深度学习是机器学习的分支，是试图使用包含复杂结构或由多重非线性变换构成多个处理层对数据进行高度抽象的算法，即根据数据进行表现特征学习的机器学习算法。将深度学习算法用于全切片数字化病理图像可能提高诊断的准确性和有效性。那么，深度学习算法与病理医生诊断相比，检测乳腺癌组织切片淋巴结转移的辨别准确性如何？
2017年12月12日，《美国医学会杂志》发表奈梅亨拉德堡德大学医学中心、埃因霍温理工大学、乌得勒支大学医学中心的人工智能诊断准确性研究报告，对自动化深度学习算法检测乳腺癌女性腋窝淋巴结清扫的淋巴结苏木精伊红染色组织全切片数码图像有否癌症转移进行评定，并与临床病理医生的诊断进行比较。
2016年研究者挑战赛（CAMELYON16）于2015年11月～2016年11月从上述两家荷兰医学中心根据免疫组织化学染色确认有无淋巴结转移的270例全切片数码图像（110例有转移、160例无转移）作为练习数据集（演算组）供挑战赛参赛者构建算法，共有390支来自世界各地的参赛队报名，其中23支参赛队在截止日期前提交了32种算法（其中25种基于深度卷积神经网络）用于检测淋巴结转移的自动化解决方案。随后通过另外129例验证数据集（验算组）全切片数码图像（49例有转移、80例无转移）对算法进行评定。同时，验算组相应玻璃切片由11位荷兰病理医生组成专家组进行有时间限制的评估，限制在2小时左右的时间内，模仿日常病理工作流程确定各个切片存在淋巴结转移的可能性，另外1位病理医生无时间限制。通过受试者操作特征（ROC）曲线下面积（AUC）对存在特定转移病灶以及玻璃切片或数码图像存在淋巴结转移与否进行定量分析。参加模拟练习的11位病理医生将其诊断把握评为绝对正常、可能正常、模棱两可、可能肿瘤、肯定肿瘤。
结果发现：
无时间限制的病理医生需要大约30小时对129例全切片数码图像进行评定，无假阳性（即非肿瘤组织被辨别为转移），但是27.6%的转移未被发现，故辨别有无病变的真阳性率为72.4%，辨别转移分类的敏感度、特异度、AUC分别为93.8%、98.7%、0.966。
有时间限制的病理医生需要72～180分钟，辨别转移分类的平均敏感度、特异度、AUC分别为62.8%、98.5%、0.810。
通过对32种算法的横断面分析，所有人工智能算法的AUC范围为0.556～0.994。
最佳算法辨别有无病变的真阳性率为80.7%，平均假阳性率为1.25%，辨别转移分类的AUC为0.994，显著高于有时间限制的病理医生（P<0.001）。
排名前5位算法辨别转移分类的平均AUC为0.960，与无时间限制的病理医生相似。
排名前7位算法辨别转移分类的AUC均高于11位病理医生。
哈佛＋麻理工二队（P<0.001）
哈佛＋麻总院三队（P<0.001）
哈佛＋麻总院一队（P<0.001）
香港中文大学三队（P<0.001）
德国慕尼黑ExBI队（P=0.02）
香港中文大学一队（P=0.04）
哈佛＋麻总院二队（P=0.04）
因此，在挑战赛环境下，通过模仿日常病理工作流程的模拟练习，一些深度学习算法对全切片数字化图像的诊断能力，优于有诊断时间限制的11位病理医生，接近无诊断时间限制的病理专家。这些结果表明了深度学习算法用于病理诊断的潜力，但是需要在临床环境下对其实用性进行评定。
对此，美国纽约纪念医院斯隆凯特林癌症中心的流行病学和生物统计学专家发表研究方法解读：使用自由反馈受试者操作特性曲线评定机器诊断癌症的准确性。
对此，美国布列根医院和波士顿妇女医院的病理学专家发表述评：深度学习算法帮助人工智能检测乳腺癌淋巴结转移。
深度学习算法帮助人工智能检测乳腺癌淋巴结转移
相关阅读
JAMA. 2017 Dec 12;318(22):2199-2210.
Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer.
Babak Ehteshami Bejnordi; Mitko Veta; Paul Johannes van Diest; Bram van Ginneken; Nico Karssemeijer; Geert Litjens; Jeroen A. W. M. van der Laak; CAMELYON16 Consortium.
Radboud University Medical Center, Nijmegen, the Netherlands; Eindhoven University of Technology, Eindhoven, the Netherlands; University Medical Center Utrecht, Utrecht, the Netherlands.
This diagnostic accuracy study compares the ability of machine learning algorithms vs clinical pathologists to detect cancer metastases in whole-slide images of axillary lymph nodes dissected from women with breast cancer.
QUESTION: What is the discriminative accuracy of deep learning algorithms compared with the diagnoses of pathologists in detecting lymph node metastases in tissue sections of women with breast cancer?
FINDING: In cross-sectional analyses that evaluated 32 algorithms submitted as part of a challenge competition, 7 deep learning algorithms showed greater discrimination than a panel of 11 pathologists in a simulated time-constrained diagnostic setting, with an area under the curve of 0.994 (best algorithm) vs 0.884 (best pathologist).
MEANING: These findings suggest the potential utility of deep learning algorithms for pathological diagnosis, but require assessment in a clinical setting.
IMPORTANCE: Application of deep learning algorithms to whole-slide pathology images can potentially improve diagnostic accuracy and efficiency.
OBJECTIVE: Assess the performance of automated deep learning algorithms at detecting metastases in hematoxylin and eosin-stained tissue sections of lymph nodes of women with breast cancer and compare it with pathologists' diagnoses in a diagnostic setting.
DESIGN, SETTING, AND PARTICIPANTS: Researcher challenge competition (CAMELYON16) to develop automated solutions for detecting lymph node metastases (November 2015-November 2016). A training data set of whole-slide images from 2 centers in the Netherlands with (n=110) and without (n=160) nodal metastases verified by immunohistochemical staining were provided to challenge participants to build algorithms. Algorithm performance was evaluated in an independent test set of 129 whole-slide images (49 with and 80 without metastases). The same test set of corresponding glass slides was also evaluated by a panel of 11 pathologists with time constraint (WTC) from the Netherlands to ascertain likelihood of nodal metastases for each slide in a flexible 2-hour session, simulating routine pathology workflow, and by 1 pathologist without time constraint (WOTC).
EXPOSURES: Deep learning algorithms submitted as part of a challenge competition or pathologist interpretation.
MAIN OUTCOMES AND MEASURES: The presence of specific metastatic foci and the absence vs presence of lymph node metastasis in a slide or image using receiver operating characteristic curve analysis. The 11 pathologists participating in the simulation exercise rated their diagnostic confidence as definitely normal, probably normal, equivocal, probably tumor, or definitely tumor.
RESULTS: The area under the receiver operating characteristic curve (AUC) for the algorithms ranged from 0.556 to 0.994. The top-performing algorithm achieved a lesion-level, true-positive fraction comparable with that of the pathologist WOTC (72.4% [95% CI, 64.3%-80.4%]) at a mean of 0.0125 false-positives per normal whole-slide image. For the whole-slide image classification task, the best algorithm (AUC, 0.994 [95% CI, 0.983-0.999]) performed significantly better than the pathologists WTC in a diagnostic simulation (mean AUC, 0.810 [range, 0.738-0.884]; P<.001). The top 5 algorithms had a mean AUC that was comparable with the pathologist interpreting the slides in the absence of time constraints (mean AUC, 0.960 [range, 0.923-0.994] for the top 5 algorithms vs 0.966 [95% CI, 0.927-0.998] for the pathologist WOTC).
CONCLUSIONS AND RELEVANCE: In the setting of a challenge competition, some deep learning algorithms achieved better diagnostic performance than a panel of 11 pathologists participating in a simulation exercise designed to mimic routine pathology workflow; algorithm performance was comparable with an expert pathologist interpreting whole-slide images without time constraints. Whether this approach has clinical utility will require evaluation in a clinical setting.
DOI: 10.1001/jama.2017.14585
JAMA. 2017 Dec 12;318(22):2250-2251.
Using Free-Response Receiver Operating Characteristic Curves to Assess the Accuracy of Machine Diagnosis of Cancer.
Chaya S. Moskowitz.
Memorial Sloan Kettering Cancer Center, New York, New York
This JAMA Guide to Statistics and Methods discusses the use of free-response receiver operating characteristic curves to test the accuracy of computer algorithms to detect the localization of disease on pathology slide images.
DOI: 10.1001/jama.2017.18686
JAMA. 2017 Dec 12;318(22):2184-2186.
Deep Learning Algorithms for Detection of Lymph Node Metastases From Breast Cancer: Helping Artificial Intelligence Be Seen.
Jeffrey Alan Golden.
Brigham and Women's Hospital, Boston, Massachusetts.
Interview with Jeffrey Alan. Golden, MD, author of Deep Learning Algorithms for Detection of Lymph Node Metastases From Breast Cancer: Helping Artificial Intelligence Be Seen.
DOI: 10.1001/jama.2017.14580
联系客服
