Overview

总览

This article is for anyone who, like me, is interested in pursuing AI research in the healthcare space. Just under a month ago, I began my Master’s program research in partnership with a local hospital. I want to give you an idea of my background, my current day-to-day work, and my initial thoughts on the field.

本文适用于像我这样对在医疗保健领域进行AI研究感兴趣的任何人。 不到一个月前，我与当地一家医院合作开始了我的硕士课程研究。 我想向您介绍我的背景，当前的日常工作以及对该领域的初步想法。

As with anything, what I have to say will almost certainly not be applicable to all AI healthcare research, but I hope you can walk away with some general concepts.

就像其他任何事情一样，我不得不说的几乎肯定不会适用于所有 AI医疗保健研究，但是我希望您可以摆脱一些一般概念。

My Background

我的背景

I am a 23 year old Master’s student at Ohio State University studying… Mechanical Engineering? And my undergraduate degree? Mechanical Engineering… so how did I get here? Let’s walk it back a bit.

我是俄亥俄州立大学的23岁硕士研究生，正在学习……机械工程？ 还有我的大学学位？ 机械工程...那我怎么来的？ 让我们往回走一点。

Since I was maybe 10, I have loved computers. I still remember telling my dad that I wanted to learn how to make video games and then coming back from school to a crisp new copy of “C++ For Dummies” sitting on my bed. I didn’t know what an integer was, what the word variable meant, or how any of this got me closer to making the spiritual successor to Star Wars Battlefront 2, but that didn’t matter. Because after spending an hour sweating over a hot keyboard, installing some ancient version of Dev-C++, I got to hit the “Run” button, and something magical happened…

从我十岁开始，我就喜欢计算机。 我仍然记得跟父亲说过，我想学习如何制作视频游戏，然后从学校回到坐在床上的新版“ C ++ For Dummies”。 我不知道整数是什么，变量一词的含义，或者说其中的任何一个使我更接近成为《星球大战前线2》的精神继任者，但这并不重要。 因为花了一个小时在热键盘上流汗，安装了一些古老版本的Dev-C ++，我才按下“运行”按钮，所以发生了一些神奇的事情……

I made my computer speak to me. Somehow, all this gibberish I didn’t understand let my computer talk. From that point on, I was almost always doing something computer related, whether it was trying to mod my favorite games or making my own basic text adventures. Much of the mystery behind computers has vanished for me, but I have always been in awe of their capabilities.

我让计算机对我说话。 不知何故，我不明白的所有这些乱七八糟的东西让我的电脑说话了。 从那时起，我几乎总是在做与计算机相关的事情，无论是尝试修改我喜欢的游戏还是进行自己的基本文字冒险。 对于我来说，计算机背后的许多谜团已经消失了，但是我一直对它们的功能敬畏。

This is my dream job, and it took a complete coincidence for me to realize it.

这是我梦dream以求的工作，而这完全是我实现的巧合。

So now that I’ve thoroughly ousted myself as a hyper-nerd, you may be wondering why, if I love computers so much, I decided to study Mechanical Engineering instead. The main reason was that I feared I wouldn’t get to work on physical problems if I became a programmer. My vision would fade from staring at a computer screen all day, I would develop horrible posture, my teeth would fall out, etc. No one warned me that all of that would be the reality of mechanical engineering as well!

因此，既然我已经彻底抛弃了自己，成为一个超级书呆子，您可能想知道为什么，如果我如此热爱计算机，却决定学习机械工程。 主要原因是我担心如果成为一名程序员，我不会从事物理问题的研究。 一整天盯着电脑屏幕，我的视力会消失，我会变得可怕的姿势，牙齿会掉下来，等等。没有人警告我，所有这些都是机械工程的现实！

But realizing those realities after an extended internship at Honda wasn’t what changed my path. A strange crossroads of my capstone project being medical in nature and my hopeful Master’s program advisor looking into artificial intelligence research with hospitals near campus led me here. To put it plainly, this is my dream job, and it took a complete coincidence for me to realize it. Focusing on programming again feels like coming home.

但是，在本田延长实习后才意识到这些现实并没有改变我的道路。 我的顶峰项目本质上是医学领域，这是一个奇怪的十字路口，而我充满希望的硕士项目顾问在校园附近医院研究人工智能研究时，把我带到了这里。 简而言之，这是我梦dream以求的工作，而实现它完全是巧合。 再次专注于编程感觉就像回到家了。

My Day-To-Day

我的日常

So now that I’ve talked about myself, what is it that I actually do? Like with most jobs, I don’t have a set list of to-dos coming from on high, but there are several things I do every day. Here’s the three most common for me, in no particular order.

因此，既然我已经谈论了自己，那么我实际上在做什么？ 像大多数工作一样，我没有固定的待办事项清单，但是每天我都会做几件事。 这对我来说是最常见的三个，排名不分先后。

1. Fiddling with Data

   摆弄数据

This is true for basically any Machine Learning job, and it can’t be overstated. Most of my days have been focused on cleaning and processing my own data or reading about how other researchers with similar data have cleaned and processed theirs. How did they handle missing values? Did they have any uniquely engineered features? How did they normalize their data? Was their model able to achieve good results with their methods? These are all questions you’ll be asking when working through the papers of your fellow researchers.

基本上对于任何机器学习工作都是如此，并且不能夸大其词。 我大部分时间都专注于清理和处理我自己的数据，或阅读有关其他具有类似数据的研究人员如何清理和处理自己的数据的信息。 他们如何处理缺失的价值？ 他们有独特设计的功能吗？ 他们如何标准化数据？ 他们的模型能否通过他们的方法取得良好的结果？ 这些都是您在研究其他研究人员的论文时会问的所有问题。

2. Working with Stakeholders

2. 与利益相关者合作

Obviously, the purpose of research is to achieve some result and share it with the scientific community. But what result? What if Stakeholder A has a different idea from Stakeholder B of what you should be doing? You need to make sure that everyone is on the same page, heading towards the same goal, and most importantly, you need to make sure that you understand where you’re headed. While your stakeholders are nominally in charge, you are likely the subject matter expert when it comes to your data. It’s critical that you inform them of any limitations or holes in the data set.

显然，研究的目的是取得一些成果并与科学界分享。 但是结果如何呢？ 如果利益相关者A与利益相关者B有不同的想法该怎么办？ 您需要确保每个人都在同一页面上，朝着相同的目标前进，最重要的是，您需要确保您了解前进的方向。 当您的利益相关者名义上负责时，就数据而言，您很可能是主题专家。 告知他们数据集中的任何限制或漏洞是至关重要的。

There is a story behind all data, but you must be able to ask it the right questions for that story to come to light. If, for example, your stakeholders want to be able to detect internal bleeding in patients, but despite your best efforts, your model can do no better than random chance, it is your responsibility to share your concerns with your stakeholders. A quick change in direction, investment in new data, or perhaps a suggestion for a feature you didn’t think to engineer are all far more useful than banging your head against the wall.

所有数据背后都有一个故事，但是您必须能够向它提出正确的问题，以使该故事浮出水面。 例如，如果您的涉众希望能够检测到患者的内部出血，但是尽管您已尽了最大努力，但是您的模型所带来的好处莫过于随机机会，因此有责任与您的涉众分享您的担忧。 方向的快速变化，对新数据的投资，或者对您不曾想过设计的功能的建议，都比将头撞墙更有用。

3. Reading Research Papers

3. 阅读研究论文

This really shouldn’t come as a surprise! Being a healthcare researcher doesn’t just mean researching your own problem. It’s almost always a good investment of your time to read the research of your peers using machine learning in healthcare. While not as populated as machine learning research in other fields (yet!), the insights of other researchers, even when they are dealing with completely different problems from my own, can often easily be applied to my own work.

这真的不足为奇！ 成为医疗保健研究人员不仅意味着要研究自己的问题。 在医疗保健中使用机器学习来阅读同行的研究几乎总是您的时间上的一笔好投资。 尽管还没有其他领域的机器学习研究那么丰富(但！)，但其他研究者的见解，即使他们正在处理与我自己的问题完全不同的问题，也常常可以轻松地应用于我自己的工作。

If you are serious about becoming a researcher, get serious about reading research. It can be very taxing, in terms of both time and effort, but the net benefit is worth it. I read at least one paper a day. Think of it like this: If the optimal solution were trivial, why would anyone research it? And if someone else has already researched it and found a good solution, why would I need to reinvent the wheel?

如果您认真对待成为研究人员，请认真阅读研究。 就时间和精力而言，这可能会非常繁重，但净收益是值得的。 我每天至少读一篇论文。 这样想：如果最佳解决方案是微不足道的，那么为什么会有人研究呢？ 而且，如果其他人已经对其进行了研究并找到了一个好的解决方案，那为什么我需要重新发明轮子呢？

If you are serious about becoming a researcher, get serious about reading research.

如果您认真对待成为研究人员，请认真阅读研究。

My Initial Thoughts on the Field

我在田野上的初步想法

Since this article has already been a decent length, I will try to keep this brief! The below are some of my hot takes from this past month.

由于本文已经足够长，因此我将尽量保持简短！ 以下是我过去一个月的一些热门话题。

1. There is a bright future for AI in healthcare! Some of its capabilities are only just being realized, and there are many exciting areas of ML study unique to health data人工智能在医疗保健领域有光明的未来！ 它的某些功能只是刚刚被意识到，并且健康数据有很多令人兴奋的ML研究领域
2. The lack of publicly available solutions to certain problems can be confounding, but I believe this will become less of a problem over time对于某些问题缺乏公开可用的解决方案可能会令人困惑，但是随着时间的推移，我相信这将不再是一个问题
3. Domain knowledge is absolutely critical. Having some health data is not enough to be a competent healthcare researcher. Working with health professionals is a must领域知识绝对至关重要。 拥有一些健康数据不足以成为合格的医疗保健研究人员。 必须与卫生专业人员合作

Conclusion

结论

Thank you for reading my article! Hopefully there will be more in the future. If you liked what you saw, consider following me on LinkedIn or even just Medium.

感谢您阅读我的文章！ 希望将来会有更多。 如果您喜欢自己看到的内容，请考虑在LinkedIn或仅在Medium上关注我。

Let me know your thoughts in the comment section! Are you interested in pursuing research in AI healthcare? Are you a researcher now? If yes, what are your thoughts on the field?

在评论部分让我知道您的想法！ 您对从事AI医疗研究感兴趣吗？ 您现在是研究员吗？ 如果是，您对此有何想法？