對沖基金分析員用深度學習診斷心臟狀況
ByCAT ZAKRZEWSKIMar
30, 2016 9:11 pm ET
Two quantitative analysts using artificial intelligence in an
online data science competition showed they could diagnose heart disease about
as accurately as doctors.
在線數據科學比賽中應用了人工智能后，兩位定量分析員發現，他們能夠像醫生一樣準確地診斷心臟病了。
Qi Liu and Tencia Lee, hedge fund analysts and self-described
“quants,” built the winning algorithm in the competition, which could find indicators
of heart disease. The online data contest challenged participants to develop
machine algorithms that could measure cardiac volumes from MRIs provided by the
National Heart, Lung and Blood Institute.
兩位自稱為“數量分析專家”的對沖基金分析員，劉秦（音）和李特西亞（音）開發了這種能夠發現心臟病的指向標獲勝算法。這次在線數據競賽要求參與者開發出一種機器算法，這種算法要能夠從美國國立心肺血液研究院（the National Heart, Lung and Blood
Institute）提供的核磁共振成像影像中測量出心容積。
Mr. Liu and Ms. Leedidn’t know each other before they won the competition, beating out more than1,390 algorithms. They met each other in a forum on the Kaggle site, where thecompetition was hosted over a three-month period.
在打敗其他超過1390算法贏得比賽前，劉先生和李女士并不認識彼此。他們是在一次Kaggle網站的論壇上遇到的，也正是Kaggle舉辦了這次為期超過三個月的數據科學比賽。
“We decided to combine our methods,” Ms. Lee said. “We decided
they were different enough from each other that we could do better than either
of us would alone.”
“我們決定把我們的方法結合起來”李女士說。“相比較而言，我們覺得我們的算法足夠與眾不同，所以如果我們兩人合作的話，效果會比獨自一個人單獨做更好。”
Competitors worked on algorithms that could accomplish a manual
and slow process that normally is carried out by cardiologists. Usually it
takes doctors about 20 minutes to measure cardiac volumes and derive ejection
fraction data from an MRI. The algorithms can analyze the images much more
quickly.
在這次比賽中，參賽選手們被要求開發出一種算法，以實現通常要心臟病專家手動且緩慢才能做到的過程。通常想要做到這些，醫生需要通過核磁共振成像技術，花費大約二十分鐘，才能檢測出心容積并得到射血分析數據。而在應用了算法以后，則可以更快分析出這些影像。
The data scientists had a set of more than 1,000 MRIs to work
with. Ms. Lee said the winning algorithm used a technique called deep learning.
在比賽中，數據科學家需要研究一組超過一千張的核磁共振成像照片。李女士說，他們在算法中是運用到了深度學習技術才得以獲勝的。
The National Heart, Lung and Blood Institute will now test the
algorithm in clinical environments. Ms. Lee said she hopes one day the
algorithm can be used by health care professionals, but she said there is a
long road of testing and regulatory processes before it gets there.
現在，美國國立心肺血液研究院正在臨床環境中測試這套算法。李女士說，她希望有以天，醫療保健專業人士能夠運用到這套算法，但是她也表示，在那一天到來之前，這套算法的測試和規范過程還有很長的路要走。
“I certainly wouldn’t trust my doctor with what we just wrote,”
Ms. Lee said.
“我肯定不會相信我的醫生只是運用了我寫的算法。”李女士說。
Many companies are trying to use artificial intelligence to
improve medicine. Investors are increasingly backing startups in the category,
and public technology companies have signaled it is a sector they’re betting
on. Last year International Business Machines Corp. acquired Merge Healthcare
Inc. for $1 billion, according to The Wall Street Journal.
近些年來，許多公司正在試圖運用人工智能去改進藥物。支持這一領域創業公司的投資者正在不斷增長，與此同時，公共科技公司也放出信號說，他們也將會把賭注押在這一領域。據華爾街日報報道，去年，IBM就以10億美金收購了醫學成像及臨床系統供應商Merge Healthcare Inc，并將其與旗下沃森健康（Watson Health）部門合并。
The contest, the National Data Science Bowl, is sponsored by the
data science startup Kaggle and Booz Allen Hamilton.
這場國家數據科學碗（the National Data Science Bowl）比賽，是由數據科學創業平臺Kaggle和博思艾倫咨詢公司（Booz Allen Hamilton）贊助的。
This is the second time Kaggle has hosted a National Data
Science Bowl. The first competition challenged scientists to come up with
algorithms that could measure plankton population to predict global health.
這是Kaggle第二次舉辦國家數據科學碗比賽。第一次比賽是要求科學家們開發一種能夠通過檢測浮游生物數量來預測全球健康的算法。
Kaggle Chief Executive AnthonyGoldbloom said the winner of the firstNational Data Science Bowl now works as a research scientist at Google’sDeepMind. He said Kaggle competitions, which crowd source solutions to big dataproblems, have become a way for data scientists to quantify their skills whenapplying for jobs.
Mr. Goldbloom said the company is now seeking proposals for the
third competition’s challenge.
Kaggle的首席執行官安東尼·古德魯姆說，第一次國家數據科學碗的獲勝選手現在正在谷歌Deepmind做研究科學家。他說，Kaggle的競賽能夠為大數據問題眾籌解決方案，這也成為數據科學家們申請工作時量化他們能力。
古德魯姆先生說，公司正在為第三次競賽尋找提案。
Write to Cat Zakrzewski at
cat.zakrzewski@wsj.com. Follow her on Twitter at @Cat_Zakrzewski.
