Google: AI Found Breast Cancer with 99 Percent Accuracy
Google says it is deeply involved in developing artificial intelligence, or AI, to help doctors identify and treat disease.
The company previously announced successful tests of machine learning systems designed to assist doctors.
In one case, Google reported AI had examined images to diagnose diabetic eye disease with equal accuracy to doctors. Other tests showed that machine learning can be used to study massive amounts of patient data to predict future medical events.
Now the company has published two new studies showing a high level of success in identifying one of the deadliest forms of cancer. Google carried out tests with researchers from the Naval Medical Center San Diego. Google reported the findings on its AI website and results were also published in scientific journals.
Using AI to detect metastatic cancer
The tests involved machine learning methods to examine digitized images of body tissue samples for signs of metastatic breast cancer. Metastatic means cancer that has spread from its main area to other parts of the body.
This kind of cancer is among the most difficult to identify and treat. Metastatic breast cancer is also one of the deadliest, causing an estimated 90 percent of all breast cancer deaths worldwide.
In metastatic breast cancer patients, the cancer often travels to nearby lymph nodes. Usually doctors examine lymph node tissue samples under a microscope to see whether cancer is present.
It is important for doctors to identify metastatic cancer as quickly as possible to prevent further spreading. The process is also important to help make decisions about treatments.
Google notes that previous studies have shown that up to one-fourth of metastatic lymph node classifications end up being changed after a second examination. In addition, studies show that small metastatic material can be missed up to 67 percent of the time when examinations happen under extreme time restrictions.
Google says it created a mathematical algorithm that greatly improves this accuracy rate. The algorithm, called Lymph Node Assistant, is trained to find characteristics of tissue affected by metastatic cancer. When the system examined tissue images, it was able to differentiate between metastatic cancer and non-cancer 99 percent of the time, Google reported.
In addition, the company said the Lymph Node Assistant was highly effective at finding the positions of the cancers. Some of these positions would be too small for doctors themselves to identify, Google added.
The research also showed that the algorithm method can reduce the usual time needed to examine a sample by about 50 percent.
The system was designed with extreme sensitivity to "exhaustively" examine every part of a tissue sample, the researchers wrote in their paper.
AI not meant to replace doctors
But Google makes clear the AI-based system is not meant to replace the work of medical professionals. Instead, it is designed to reduce the number of false identifications and help doctors work faster and more effectively.
The studies suggest that "people and algorithms can work together effectively to perform better than either alone," Google said in a statement.
The company says that clearly more AI research and experiments will be necessary to further progress in the fight against breast cancer. But it is hopeful that "deep learning technologies and well-designed clinical tools" can be continually developed to improve accuracy and availability for patients worldwide.
I'm Bryan Lynn.
1. be involved in 卷入的；参与的；与…有密切关联的；
She has been involved in the war against organised crime.
2. carry out 实施；执行；实行；
The operator has to be able to ca rry out routine maintenance of the machine.
3. end up 最终；结果；到头来；
He never dreamed that he would end up owning such a lot of property.
4. differentiate between 区分；区别；辨别；
A child may not differentiate between his imagination and the real world.
谷歌表示，在其中一个案例中，人工智能通过检测图像来诊断糖尿病引起的眼病，其准确率与医生一样 。其他测试显示，机器学习可以用来研究大量的患者数据，以预测未来的医疗事件 。
现在，谷歌公司公开的两项新研究结果表明，人工智能在确定其中一种最致命癌症上的成功率很高 。谷歌和圣地亚哥海军医学中心的研究人员共同进行了测试 。谷歌表示，研究结果公布在其人工智能网站和科学期刊上 。
测试用机器学习方法来检测人体组织样本的数字图像，以寻找转移性乳腺癌征兆 。转移意味着癌症已从原发部位扩散到身体其他部位 。
这种癌症是最难确认和治疗的癌症之一 。转移性乳腺癌也是最致命的癌症之一，在全世界乳腺癌死亡病例中，90%死于癌转移 。
对转移性乳腺癌患者来说，癌症通常会转移至临近的淋巴结 。通常，医生会在显微镜下检查淋巴结组织样本，观察癌症是否已转移 。
对医生来说，尽快检测出转移性癌症并阻止癌症进一步扩散非常重要 。这一过程对于决定治疗方案来说同样重要 。
谷歌指出，此前的研究表明，有四分之一的淋巴结转移分类会在第二次检查之后发生改变 。另外，研究还表明，在极端时间限制下进行的检查，没能发现小转移的比例高达67% 。
谷歌表示，他们研发的一种数学算法极大地提高了检测的准确率 。这种算法名为“淋巴结助手”，它经过了训练，目的是找出被转移性癌症影响的组织特征 。谷歌表示，系统检测组织图像时能区分转移性癌症和非癌症，准确率高达99% 。
另外，谷歌还表示，淋巴结助手在确定癌症位置方面也非常有效 。谷歌说，一些位置可能非常小，医生无法识别 。
但是谷歌明确表示，这个基于人工智能的系统不是为了取代医务人员的工作 。相反，其目的是减少错误识别，并帮助医生提高工作效率 。
谷歌表示，就在防治乳腺癌方面取得进一步进展来说，显然更多的人工智能研究和实验是必需的 。但是他们希望“深度学习技术和设计完善的临床工具”能不断发展，为全世界患者提高诊断的准确度和可用性 。