|
|
<section><section><p><p><img src="image/20201014/86b79d371e9862b01189f0252a03b2cd_1.png" /></p></p></section><section><section><strong></strong></section><p><span>医疗器械第一新媒体</span></p><p><span>分享最专业的医疗器械知识</span></p></section><section><section><section>关注</section></section></section></section><section data-tools="gulangu" data-id="86447" data-color="rgb(0, 91, 172)" data-custom="rgb(0, 91, 172)"></section><p><span>转自鼎湖影像</span></p><p><span><span><strong><span><br /></span></strong></span></span></p><p><span><span><strong><span>On the cover: In this issue (1122–1131), Kermany, Zhang et al. </span></strong></span><span>apply a deep-learning framework to develop a diagnostic tool for the classification and diagnosis of human diseases. The cover image depicting an iris and pupil is in fact a composite mosaic of 315 optical voherence tomography images of treatable blinding retinal diseases that were used to evaluate the performances of the image-based learning framework developed in the paper. This image was designed and composed by Daniel S. Kermany.</span></span></p><p><span><br /></span></p><p><span>正月里传来好消息，最新一期《Cell》以封面文章形式刊登了以广州妇女儿童医疗中心张康教授为Lead Contact的论文，文章题目为：<span>Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learn</span>ing。<strong>这是</strong><strong>中国研究团队首次在如此重量级医学期刊发表有关医学人工智能的文章。</strong></span></p><p><span></span></p><p><p><img src="image/20201014/3e2bea4e0997da9431ea332a1e8f7226_2.jpg" /></p></p><p><span><strong><span>Summary</span></strong></span></p><p><span>The implementation of clinical-decision support algorithms for medical imaging faces challenges with reliability and interpretability. Here, we establish a diagnostic tool based on a deep-learning framework for the screening of patients with common treatable blinding retinal diseases. Our framework utilizes transfer learning, which trains a neural network with a fraction of the data of conventional approaches. Applying this approach to a dataset of optical coherence tomography images, we demonstrate performance comparable to that of human experts in classifying age-related macular degeneration and diabetic macular edema. We also provide a more transparent and interpretable diagnosis by highlighting the regions recognized by the neural network. We further demonstrate the general applicability of our AI system for diagnosis of pediatric pneumonia using chest X-ray images. This tool may ultimately aid in expediting the diagnosis and referral of these treatable conditions, thereby facilitating earlier treatment, resulting in improved clinical outcomes.</span></p><p><span>“可靠性”和“可解释性”是进行临床决策支持算法面临的难点。在此，我们建立了一种基于深度学习框架的诊断工具，用于筛查患有常见可治的致盲性视网膜疾病的患者。我们的框架利用了迁移学习，通过传统方法的一小部分数据进行神经网络学习。将这种方法应用于光学相干断层摄影的数据集，在识别年龄相关的黄斑变性、糖尿病黄斑水肿上与相关医学专家水平相仿。通过突出神经网络识别的区域，我们提供了一种更加易懂、可解释的诊断方法。应用该人工智能系统，我们可以在胸部X线图像诊断小儿肺炎，以证明该系统的普遍适用能力。该工具可能最终有助于加速有关可治疗性疾病的诊断，有益于早期治疗，从而改善临床的结果。</span></p><p><span><br /></span></p><p><span><span><strong><span>研究人员应用了一种有效的“迁移学习算法”（“transfer learning algorithm”），如下图</span></strong></span><span></span></span></p><p><p><img src="image/20201014/451ccba159f2b373ad299a01bd5e8fd9_3.jpg" /></p></p><p><span><span>卷积神经网络的原理图</span><span></span><br /></span></p><p><br /></p><section data-mpa-template-id="341391" data-mpa-color="#ffffff" data-mpa-category="body"><blockquote mpa-from-tpl="t"><p><span>知识点：什么是迁移学习？</span></p><p><span>其实是机器学习中有一种特殊的类型。简单来讲，将先前领域或任务中学到的知识或技能应用到新的领域或任务中，即为迁移学习。相比于传统的深度学习模型，迁移学习模型所需的数据量极少。</span></p></blockquote><p><br /></p></section><p><p><img src="image/20201014/24e7b2dd769535b87e1e6061d4845d2d_4.jpg" /></p></p><p><span>工作流程图</span></p><p><span></span></p><p><p><img src="image/20201014/feb06446548fc7dc0201c8a4bdaa4876_5.jpg" /></p></p><p><span><br /></span></p><p><span>该AI模型在<span>鉴别这类眼部疾病</span>的准确性96.6%，灵敏性97.8%，特异性97.4%，ROC曲线下面积99.9%。</span></p><p><br /></p><p><p><img src="image/20201014/f013835a9782d6c65e061dfc6f9f3126_6.jpg" /></p></p><p><p><img src="image/20201014/7cb5c48eabf06e98ce18f4da5fb0ebc7_7.jpg" /></p></p><p><br /></p><p><span>该AI系统用于胸部X线图像诊断小儿肺炎的研究：</span></p><p><span>选取5,232 例小儿胸部X线图像, 其中3,883 例标记为肺炎 (2,538 例细菌性肺炎l and 1,345 例病毒性肺炎) ，1,349 例标记为正常。</span></p><p><span>在鉴别正常与肺炎组，准确性 92.8%, 敏感性93.2% ，特异性 90.1%。曲线下面积 96.8% (上图E). </span></p><p><span>鉴别细菌性肺炎与病毒性肺炎组，准确性90.7%, 敏感性 88.6% ，特异性90.9% 。（上图C 、D）.曲线下面积94.0%% (上图F).</span></p><p><br /></p><p><p><img src="image/20201014/2a04fcda197f9b97089edac7546c6c68_8.jpg" /></p></p><p><span>张康教授</span></p><p><span><strong><span>简介：</span></strong></span></p><p><span>1984四川大学生化专业毕业，之后就在哈佛和麻省理工学院念了三个博士学位。是新中国成立以来第一个在哈佛和MIT拿到三个博士学位的人。国家第三批“千人计划”入选者，教育部长江学者。现任美国加州大学圣地亚哥分校人类基因组医学研究所所长，医学和人类遗传学终身正教授。美国最佳眼科医生，世界百名眼科权威。任北京大学、中山大学、四川大学等大学客座教授。在分子遗传学、眼科学、肿瘤学以及精准医学领域拥有较高造诣，具有丰富临床实践经验和基础科研成果，是美国约翰·霍普金斯大学威尔玛眼科中心第一位来自中国大陆的眼科住院医生。曾经在美国犹他大学完成了临床医生训练，在约翰?霍普金斯大学、克里夫兰医院、犹他大学从事临床和基础科研工作。美国眼科学会、黄斑学会、美国临床研究委员会、美国视网膜和玻璃体学会、美国科学促进会等的会员。担任Nature、Science、Cell、JCI、New England Journal of Medicine, Lancet等多家杂志审稿专家。</span></p><section data-id="1658"><section><section><section data-id="1658"><section><section><section data-id="1658"><section><section><section data-id="1658"><section><section><p><span><strong>相关阅读</strong></span></p></section><p><p><img src="image/20201014/ec237188ae1e9c03eb4d9814f31b18ab_9.gif" /></p></p><p><span>国内首家“AI＋医疗”医院正式挂牌运行</span></p><p><br /></p><p><span>斯坦福101页年度AI报告：人工智能全面逼近人类能力</span></p><p><br /></p><p><span>未来50%的人将失业，而这21种奇怪的工作却是刚需！</span></p></section></section><p><br /></p></section></section></section></section></section></section></section></section></section><p><p><img src="image/20201014/ddc723b50bcaa3c373cb35a0033daacd_10.jpg" /></p></p>
|
|
发表于 2020-10-14 17:07:05