创新背景

当宠物去动物医院看病时，兽医会以段落的形式记录下来。宠物电子健康记录没有系统或广泛的基础设施。虽然手工记录的笔记可以很好地记录一次就诊，但在一家诊所，它限制了数据的使用和共享。

创新过程

邹市明和他的团队设计了一个植根于人工智能的解决方案DeepTag。DeepTag是一种算法，它可以从兽医那里读取打印好的记录，并预测动物可能患有的特定疾病。它把医疗记录的段落浓缩成代表某些疾病、症状或疾病的代码。

邹市明表示，人工智能理解和应用自然语言的能力已经取得了巨大的进步。人工智能在理解人类语言和对它们做出反应方面做得更好，利用这一进展构建算法，可以扫描整个段落，真正阅读临床记录，并解释每个词。研究人员并没有明确地告诉算法哪些词汇与哪种疾病有关。相反，它是寻找与特定诊断相关的关键词。

在训练算法时，邹市明与科罗拉多州立大学兽医学院合作，那里的一组兽医专家注释了超过10万份临床记录，为每个病例分配了疾病代码。聂利用这些数据集来“教”算法与特定疾病配对的笔记类型。然后，该团队通过测试从私人兽医办公室收集的宠物临床数据，进一步验证了算法的准确性。

一种药物在进行人体临床试验之前，通常要在小鼠或大鼠身上测试其有效性和安全性。但小型啮齿动物的生物学特性可能与人的生物学特性大不相同。狗的体型更大，在某些方面更能反映人类的生物学特性，一旦这种假想的治疗通过了“啮齿动物阶段”，它可以更准确地指出人类对治疗的反应。

狗是使用DeepTag记录的大多数患者，它们是科学家为人类开发的许多药物的很好的候选者。同样地，就像生病的人一样，有时缺乏治疗宠物疾病的批准选项，临床试验将是他们康复的最佳选择。但到目前为止，还没有什么基础设施来密切关注动物对新疗法的反应。

创新价值

DeepTag可以读取兽医打印的记录，并预测动物可能患有的特定疾病。DeepTag将允许兽医跟踪宠物疾病的流行情况，未来还可能成为跟踪动物临床试验的工具。

创新关键点

人工智能在理解人类语言和对它们做出反应方面做得更好，利用这一进展构建算法，可以扫描整个段落，真正阅读临床记录，并解释每个词。

Use artificial intelligence to help veterinarians write notes

Zou and his team designed DeepTag, an AI-rooted solution. DeepTag is an algorithm that can read printed records from veterinarians and predict specific diseases an animal may have. It condenses passages of medical records into codes that represent certain diseases, symptoms, or illnesses.
Zou said great progress has been made in AI's ability to understand and apply natural language. Ai has gotten better at understanding human speech and responding to them, using this progress to build algorithms that can scan entire paragraphs, actually read clinical notes, and interpret each word. The researchers did not explicitly tell the algorithm which words were associated with which disease. Instead, it's looking for keywords associated with a particular diagnosis.
In training the algorithm, Zou worked with Colorado State University's College of Veterinary Medicine, where a team of veterinary experts annotated more than 100,000 clinical records, assigning disease codes to each case. Nie uses these datasets to "teach" the types of notes the algorithm pairs with specific diseases. The team then further verified the accuracy of the algorithm by testing pet clinical data collected from private veterinary offices.
Before a drug can be tested in human clinical trials, its efficacy and safety are usually tested in mice or rats. But the biology of small rodents can be very different from the biology of humans. Dogs are larger and in some ways more reflective of human biology, and once this hypothetical treatment passes the "rodent stage," it could be a more accurate indication of how humans will respond to it.
Dogs are the majority of patients recorded using DeepTag, and they are good candidates for many of the drugs scientists are developing for humans. Likewise, as with sick people, there is sometimes a lack of approved options for treating pet diseases, and clinical trials would be their best option for recovery. But until now, there has been little infrastructure to keep tabs on how animals respond to new treatments.