IBM to identify cancerous cells technology breakthrough-IT information
IBM recently gratifying achievements in medicine, action continued. Following the success of a picture after diagnosis of skin cancer, IBM has released the latest achievements of the Institute said they used deep learning and neural networks in the identification of cancer cell mitosis have made huge progress on. When doctors diagnose cancerous cells, mainly through biopsies method is used to analyze tissue samples of patients. However, even if these organizations are sometimes as small as needles, pathologists need to detect signs of tumor cells disappeared, also want to observe the important characteristics of cancer cells to help doctors prescribe.
Pathologist in the analysis of samples, some tissue samples will be marked with a coloring reagent solution. Results show that the color depth of reagents and their distribution in the tissues, can differentiate between types of diseases and the deterioration of disease.
2016 tumor proliferation of breast cancer cells of the assessment challenge training samples
Pathologists then studied under a microscope that is labeled tissue samples. However long and heavy workloads in this phase, researchers will still have to deal with the percentage of samples, and that long periods of high load working will inevitably cause a decline in accuracy of diagnosis.
With the development of modern medical imaging technology and advanced learning, a pathologist who needed the assistance of computer technology, computer scientists but also for their tireless efforts. In order to verify the application of artificial intelligence technology in the medical field, scientists organized a hacker Marathon challenge.
A couple of weeks ago, at the University Medical Center Utrecht, aiyinhuofen Technical University, Beth Israel Deaconess Medical Center and Harvard Medical School with the support of the organizers in Greece Athens organized a "tumour assessment challenge" (Tumor Proliferation Assessment Challenge,TPAC), as a branch of the 2016 MICCAI Conference on international activities.
159 teams from around the world, in the campaign raced downloads on the first day of medical school to provide 500 images of breast cancer cells. As the training samples, the data sets more than 50000*50000 megapixel resolution. Indeed, this challenge was a fierce battle, until the end of the Bell, only 14 teams have submitted results.
One team member from IBM Switzerland laboratory and IBM Brazil laboratory. This international team of talent by the French, Hungary and Greece people, participate in this sector "mitosis of based on adaptive algorithm testing a" challenge. Race for months, experiences for the whole summer, but eventually to pay a return. They received the second place in the competition, with the first name only 0.004 points.
IBM researcher Erwan Zerhouni, Maria Gabrani and David Lanyi used deep learning and neural networks to solve problems in cancer
"Artificial identifying cells mitosis is an extremely tough job, in that case, it would be handed over to the computer to solve it", David Lanyi said. He previously worked in IBM at the Zurich Institute of technology in the field of deep learning-related research.
"In July this year, we started by deep learning algorithm based on neural network for the characteristics of the tissue samples for training. Training's main job is to look for the nuances of negative and positive specimens. After a period of training, machine learning is effective. ”
Erwan Zerhouni noted that "five years ago it was almost an impossible task. At present, the algorithm for diagnosis a 5600*5600 picture an hour, in a follow-up study, we can continuously optimize, to compress the time cost to less than 20 seconds, while diagnosis of any type of cancer. ”
"We try to combine the MICCAI deep learning of the latest technology to meet with Genomics data (including genomics and proteomics), in-depth analysis, in order to provide patients with more precise medical diagnosis. "From IBM Brazil lab and participated in the challenge of Matheus Viana often speculating on the future development of this project. Meanwhile, the team is preparing shared with researchers at the IBM Haifa Labs image analysis for breast cancer research.
Cancer only of IBM Corporation in the field of medical imaging studies of disease. IBM member Dr Tanveer Syeda-Mahmood, an expert in the field of medical image understanding and progress to the team plan, and plans to collaborate with research depth method introduced medical screening area. This research useful for Radiology and cardiology. Similarly, in the research of vision fatigue, pharmacology, pathology experts often have similar challenges. Syeda-Mahmood research results on display at next week's annual meeting of the Radiological Society of North America.
IBM识别癌变细胞技术取得重大突破 - IT资讯
IBM最近在医学领域成果喜人，动作不断。继成功用照片诊断皮肤癌后，IBM研究院日前发布最新成果称，他们采用了深度学习和神经网络，在识别癌变细胞的有丝分裂上取得了巨大进展。医生在诊断癌变细胞时，主要通过用活组织切片检查法分析病人组织样本的方式。然而即使这些组织有时如针头般微小，病理学家需要从中检测出肿瘤细胞消失的种种迹象，也要观测出癌变细胞出现的重要特征，以帮助医生对症下药。
病理学家在分析样本时，会将一些典型的组织样本用试剂溶液进行着色标记。结果显示，试剂颜色的深浅及其在细胞组织内的分布情况，能够区分疾病的种类及疾病的恶化程度。
2016年肿瘤扩散评估挑战赛的乳腺癌细胞训练样本
病理学家随后要在显微镜下研究这种被标记的组织样本。然而此阶段耗时长且工作量巨大，研究人员每天都要处理上百份样本，而这样长时间的高负荷工作难免会导致诊断正确率的降低。
随着近代医学影像技术及深度学习的发展，病理学家们亟需计算机技术的援助，而计算机科学家们也在为之不懈努力。为了验证人工智能技术在医疗领域中的应用效果，科学家们组织了一场黑客马拉松挑战赛。
几个星期前，在乌德勒支大学医学中心、艾因霍芬技术大学、贝斯以色列女执事医疗中心和哈佛医学院的支持下，主办方在希腊雅典进行举办了“肿瘤扩散评估挑战赛”（Tumor Proliferation Assessment Challenge，TPAC），作为2016年的MICCAI国际会议的一个分会活动。
来自全世界各地的159支团队，在活动开展的首日争分夺秒地下载医学院提供的500张乳腺癌细胞图像。作为训练样本，该数据集超过了50000*50000像素的分辨率。诚然，这场挑战赛是一场鏖战，直到比赛结束的钟声敲响，也只有14支队伍提交了结果。
其中一支队伍来自IBM瑞士实验室和IBM巴西实验室。这支藏龙卧虎的国际队伍由法国人，匈牙利人和希腊人组成，共同参与了这界“基于自适应算法的有丝分裂检测难题”挑战赛。竞赛长达数月，经历了整整一个夏天，但是付出终有回报。他们在本次比赛中一举获得第二名，与第一名只差了0.004分。
IBM研究员Erwan Zerhouni、Maria Gabrani与David Lanyi使用深度学习与神经网络解决癌症中的难题
“人工辨认细胞的有丝分裂是一个极其棘手的工作，既然如此，那就交给计算机来解决吧”，David Lanyi如是说。他在IBM工作之前曾在苏黎世理工学院从事深度学习领域的有关研究。
“在今年7月，我们开始通过基于神经网络的深度学习算法进行对组织样本的特性进行训练。训练的主要工作是寻找阴性和阳性组织样本的细微差别。在经过一段时间的训练后，机器学习的效果显著。”
Erwan Zerhouni提道，“在五年前这几乎是一项不可能完成的任务。目前，算法诊断一幅5600*5600的图片需要一个小时，在后续的研究中我们可以不断对其进行优化，从而将时间成本压缩到20秒以内，同时可以诊断任一种类型的癌症。”
“我们设法结合MICCAI最新的深度学习技术来一起迎接针对组学数据（包括基因组学和蛋白质组学）的深入分析，为病人提供更精准的医学诊断。”来自IBM巴西实验室并参与了这一挑战赛的Matheus Viana时常在思索这一项目的未来发展。同时，这支团队正准备与IBM海法实验室的研究人员共享乳腺癌成像分析的研究结果。
癌症仅仅是IBM公司在医疗图像领域研究的一类疾病。IBM会员兼医学图像领域专家Tanveer Syeda-Mahmood博士了解到该团队的进展计划，并计划与其协作研究，意将深度学习方法引入医学筛研究领域。这对放射学与心脏病学的研究颇有裨益。类似地，在视觉疲劳的研究中，药理学、病理学的专家们经常遇到类似的挑战。Syeda-Mahmood的研究成果将在下周北美放射学会的年会上展出。