超声显微镜_利用人工智能帮助通过显微镜诊断癌症

超声显微镜

技术 (Technology)

Doctors often recommend a biopsy after a physical exam or imaging has identified possible cancer. A biopsy removes tissue from the area of concern. A specialist (pathologist) then examines this material. Today, we look at the biopsy process, before turning to an exciting line of exploration: How can we use artificial intelligence to optimize our ability to diagnose cancer rapidly and accurately?

d octors经常推荐活检体检或成像已经确定了可能的癌症之后。 活检可将组织从关注区域移开。 然后由专科医生(病理学家)检查该材料。 今天，我们将探讨活检过程，然后转向激动人心的探索路线：我们如何利用人工智能来优化我们快速，准确诊断癌症的能力？

癌症诊断基础 (Cancer diagnosis basics)

To definitively diagnose cancer (for most cancer types), we typically need a biopsy. Biopsy types include:

为了明确诊断癌症(对于大多数癌症类型)，我们通常需要进行活检。 活检类型包括：

• Incisional biopsy (removal of a sample of the concerning tissue)

  切口活检 (切除相关组织的样本)

• Needle biopsy (removal a sample of concerning tissue with a needle)

  针头活检 (用针取出有关组织的样本)

• Excisional biopsy (removal of the entire area of concern)

  切除活检 (去除整个关注区域)

On a side note, cancer care is becoming increasingly personalized. The biopsy tissue needs to provide enough material not only for diagnosis but often for genetic analysis as well.

附带说明，癌症护理正变得越来越个性化。 活检组织不仅需要提供足够的材料用于诊断，而且还经常需要提供基因分析。

After the biopsy, the removed sample goes to a pathologist. This doctor analyzes the cells under a microscope. She then determines whether the tissue is benign (not cancer) or malignant (cancer).

活检后，取出的样本将交给病理学家。 这位医生在显微镜下分析细胞。 然后，她确定组织是良性的(不是癌症)还是恶性的(癌症)。

A cancer diagnosis has some subjectivity

癌症诊断有一定主观性

Some of the determination about whether cancer is present is subjective. For example, there is a form of very early-stage breast cancer known as ductal carcinoma in situ or DCIS. When the cells are more aggressive (higher grade), the expert pathologist often has no trouble rendering the correct diagnosis.

关于是否存在癌症的一些确定是主观的。 例如，存在一种非常早期的乳腺癌，称为原位导管癌或DCIS。 当细胞更具侵略性(更高等级)时，专家病理学家通常不会在进行正确诊断时遇到麻烦。

If the grade is low, diagnosis can be more challenging. Breast cancer is a continuum, with normal cells becoming hyperplastic (normal-appearing, but with too many cells present), atypical (abnormal, but not yet meeting the criteria for DCIS or cancer), ductal carcinoma in situ, and invasive cancer.

如果等级低，诊断可能会更具挑战性。 乳腺癌是一个连续过程，正常细胞增生(外观正常，但存在过多细胞)，非典型细胞(异常，但尚未达到DCIS或癌症的标准)， 原位导管癌和浸润性癌。

Ductal carcinoma in situ (DCIS). Photomicrograph courtesy of the author.

导管原位癌(DCIS)。 显微照片由作者提供。

Because low-grade DCIS is not so different from atypia, mistakes can happen. These errors can be quite consequential, as we typically treat DCIS, but not atypical ductal hyperplasia or ADH.

由于低品位DCIS与非典型性差异不大，因此可能会发生错误。 这些错误可能是相当严重的，因为我们通常会治疗DCIS，而不是非典型的导管增生或ADH。

So how often do pathologists make errors when it comes to a breast cancer diagnosis? It appears that they are quite good at determining when invasive breast cancer is present. On the other hand, they struggle more with less severe conditions or when the tissue is normal.

那么，病理学家在诊断乳腺癌时经常犯错误的几率呢？ 看来他们很擅长确定何时存在浸润性乳腺癌。 另一方面，它们在较不严重的情况下或组织正常时会更加挣扎。

Pathologists are human and can make errors in diagnosis

病理学家是人类，可以在诊断中出错

A study of over one hundred pathologists in the USA compared the doctors’ diagnoses against those of three national experts. Publishing in 2015, researchers discovered:

一项针对美国一百多位病理学家的研究将医生的诊断与三位国家专家的诊断进行了比较。 研究人员在2015年发表的论文中发现：

• Pathologists correctly diagnosed abnormal, pre-cancerous cells about half the time. Management usually involves monitoring and sometimes risk-reducing medicines.
  病理学家正确地诊断出异常的癌前细胞约有一半的时间。 管理通常涉及监测药物，有时还涉及降低风险的药物。
• Roughly one-third of these cases were misdiagnosed as not problematic, while 17 percent were deemed more suspicious or cancer.
  这些病例中大约有三分之一被误诊为没有问题，而17％被认为更可疑或患有癌症。

Pathologists mistakenly noted something suspicious in 13 percent of normal tissue. Looking more specifically at DCIS:

病理学家错误地注意到13％的正常组织中有可疑物质。 更具体地看一下DCIS：

• Pathologists misdiagnosed 13 percent of these cases as less severe, while 3 percent were mistaken for invasive cancer.
  病理学家将这些病例中的13％误诊为较轻，而3％则误诊为浸润性癌症。

The researchers did not have information on patient outcomes, so we don’t know if the experts made the right diagnosis. In the study, pathologists could not consult with their colleagues; they routinely do in my institution. If there is no clear consensus, we send the biopsy material to an international expert.

研究人员没有有关患者预后的信息，因此我们不知道专家是否做出了正确的诊断。 在这项研究中，病理学家无法与同事协商。 他们通常在我的机构里做。 如果没有明确的共识，我们会将活检材料发送给国际专家。

National Cancer Institute on 美国国家癌症研究所的 Unsplash 照片

走向未来 (Into the future)

Recognizing the approximately 5 percent of cancer diagnoses are incorrect, The United States Department of Defense recently awarded a contract to Google to improve diagnosis accuracy. The technology company is already developing neural networks, using the Google Cloud Healthcare API to train an artificial intelligence program.

认识到大约5％的癌症诊断不正确，美国国防部最近与Google签订了一项合同，以提高诊断准确性。 这家技术公司已经在开发神经网络，使用Google Cloud Healthcare API来训练人工智能程序。

The biopsy material is de-identified. Once Google trains the AI, the company looks forward to designing a microscope. The device will have an augmented reality overlay to show the reading pathologist information about the cells’ probability of being malignant or cancer.

活检材料被取消标识。 Google训练完AI后，该公司期待设计显微镜。 该设备将具有增强现实覆盖图，以显示正在读取的病理学家有关细胞发生恶性或癌症可能性的信息。

If successful, this approach would join others that are in current use in cancer medicine. Your pathologist may employ artificial intelligence to see if breast cancer has an antennae-like protein known as HER2. Your radiologist may use AI to provide an additional read for your mammograms. The future is coming fast.

如果成功，这种方法将与目前在癌症医学中使用的其他方法结合在一起。 您的病理学家可能会利用人工智能来检查乳腺癌是否具有称为HER2的触角样蛋白。 您的放射科医生可能会使用AI为您的乳房X线照片提供额外的读数。 未来很快。

I’m Dr. Michael Hunter. Thank you for joining me today.

我是Michael Hunter博士。 谢谢您今天加入我的行列。

翻译自: https://medium.com/beingwell/using-artificial-intelligence-to-help-diagnose-cancer-by-microscope-c212fae2570b

超声显微镜