ADM-201 dump PMP dumps pdf SSCP exam materials CBAP exam sample questions

DeepMind的人工智能眼疾诊断系统 – 译学馆
未登录,请登录后再发表信息
最新评论 (0)
播放视频

DeepMind的人工智能眼疾诊断系统

DeepMind's New AI Diagnoses Eye Conditions

亲爱的学者们 这是Karoly Zsolnai Feher2分钟时长的论文
Dear Fellow Scholars, this is Two Minute Paperswith Károly Zsolnai-Fehér.
在系列视频中
In this video series,
我们经常看到这些神奇的机器学习算法
we often see how these amazing new machine learning algorithms can
使我们的生活更方便
make our lives easier,
幸运的是 有一些还应用到医疗方面
and fortunately, some of them are also useful for serious medical
特别是
applications. Specifically,
医疗成像
medical imaging.
医疗成像普遍用于
Medical imaging is commonly used
医疗保健系统 把一套挑选出来的组织或器官的图像
in most healthcare systems where an image of a chosen set of
提供给医生看
organs and tissues is made for a doctor to look
来决定是否有必要
at and decide whether medical intervention
医疗介入
is required.
主要问题是
The main issue is
现在诊断的图像正在以
that the amount of diagnostic images out there in the wild increases at
一个惊人的速度增加
a staggering pace,
这就使得医生越来越看不过来
and it makes it more and more infeasible for doctors to look at.
但 稍等
But wait a minute,
图像越来越多 同时也就意味着
as more and more images are created, this also means that we have
我们有更多的训练数据提供给机器学习算法
more training data for machine learning algorithms,
因此 当医生淹没在越来越多的图像中
so at the same time as human doctors get more
经过一段时间 人工智能会变得越来越厉害
and more swamped, the AI should get better and better over time!
这些方法通过多方位的图像形成结论
These methods can process orders of magnitude more
这种算法比人有优势 接下来
of these images than humans, and after that,
最终的结论
the final decision is put back
交到医生手里
into the hands of the doctor, who can now focus
医生去关注那些极端案例 优先考虑哪些病例要立即处理
more on the edge cases and prioritize which patients should be seen immediately.
这个来自深思项目科学家的工作
This work from scientists at DeepMind was trained
被输入了1.4万张连续图像的
on about 14 thousand optical coherence
X射线断层扫描图片 这是OCT标签
tomography scans, this is the OCT label you see
你看左边
on the the left, these images are cross
这些图像是人类的视网膜切片
sections of the human retina.
我们先从OCT扫描开始
We first start our with this OCT scan, then,
接下来跟进的是手工挑选步骤
a manual segmentation step follows, where
医生标注图片 标明和哪个部分相关
a doctor marks up this image to show where the most relevant parts,
像这个视网膜
like the retinal
是关于视网膜色素的体液或位置的
fluids or the elevations of retinal pigmentsare.
在得出结论之前 暂停一会儿
Before we proceed, let’s stop here
看这些图片是怎样通过网络
for a moment and look at some images of how the network
向医生学习 最后完成归类的
can learn from the doctors and reproduce thesesegmentations by itself.
看那个
Look at that!
图像清晰
It’s almost pixel perfect!
似乎是科幻一般
This looks like science fiction.
现在我们有了归类图 是时间表演一下分级了
Now that we have the segmentation map, it is time to perform classification.
就是说
This means that we look
我们看看这个图 可能分配给每一个可能出现的状况
at this map and assign a probability to each possible condition that
最后
may be present. Finally,
根据这些
based on these,
做出最终结论 病例是否要立即处理
a final verdict is made whether the patient needs to be urgently seen,
或者只做例行检查 或者连检查都没必要
or just a routine check, or perhaps no check is required.
算法也要学习这个分类的步骤 才能得出最后的结论
The algorithm also learns this classification step and creates these verdicts itself.
当然了 问题来了 这样得出的结论准确吗?
And of course, the question naturally arises:how accurate is this? Well,
我们来看看这个令人迷惑的模型
let’s look at the confusion matrices!
模型告诉我们
A confusion matrix shows us how many
有多少病例被正确地分类了出来 从中分类出急诊病例
of the urgent cases were correctly classified as urgent,
而有时被错误分类为
and how often it was misclassified
其他的事 到底是什么
as something else and what the something else
其他事
was.
诸如此类的分析应用到其他类的分级
The same analysis is performed to all otherclasses.
这些是视网膜专家做的
Here is how the retina specialist doctors did,
这些是AI做的
and here is how the AI did.
我在那里稍停片刻 你们可以比较一下
I’ll leave it there for a few seconds for you to inspect it.
真的不错
Really good!
这是另一种汇集起这些数据的方法
Here is also a different way of aggregating this data –
算法干的很出色
the algorithm did significantly
比所有的验光师还有排名第一的视网膜专家
better than all of the optometrists and matched the performance
都要好
of the number one retina specialist.
假如不是我亲眼所见这份报告
I wouldn’t believe any of these results
我不愿相信
if I didn’t see these reports with my own eyes
任何的结果
in the paper.
这项技术另外的优势在于
An additional advantage of this technique is
它可应对不同的图片设备
that it works on different kinds of imaging
第一种方法中 它能应对3D数据
devices and it is among the first methods that works with 3D data.
另外 我特别喜欢的是
Another plus that I really liked is that this was developed
它是和一所英国伦敦顶尖眼科医院
as a close collaboration with
共同合作开发的
a top tier eye hospital in London to make sure
来确保结果尽可能的
that the results are as practical as
有效
possible.
论文中包含了大量的信息
The paper contains a ton of more information,
一定要看看
so make sure to have a look!
这是深思项目历经艰苦后取得的
This was a herculean effort from the side of DeepMind,
惊人的科学成果
and the results are truly staggering.
多有生气啊
What a time to be alive!
谢谢观影 感谢你无私的支持
Thanks for watching and for your generous support,
下次见
and I’ll see you next time!

发表评论

译制信息
视频概述

不可估量

听录译者

收集自网络

翻译译者

Tiger_woo

审核员

审核员 W

视频来源

https://www.youtube.com/watch?v=wR2OlsF1CEY

相关推荐