字幕列表 影片播放 列印所有字幕 列印翻譯字幕 列印英文字幕 Computer algorithms today are performing incredible tasks 譯者: Lilian Chiu 審譯者: Helen Chang with high accuracies, at a massive scale, using human-like intelligence. 現今的電腦演算法能夠執行 很了不起的工作任務, And this intelligence of computers is often referred to as AI 有高度的精確性,規模可以很大, 且用的是類似人類的智慧。 or artificial intelligence. 這種電腦的智慧通常被稱為 AI, AI is poised to make an incredible impact on our lives in the future. 也就是人工智慧。 Today, however, we still face massive challenges 人工智慧已經準備好要對 我們未來的生活造成衝擊。 in detecting and diagnosing several life-threatening illnesses, 然而我們現今仍然面臨很大的挑戰, such as infectious diseases and cancer. 包括偵測與診斷數種 會威脅生命的疾病, Thousands of patients every year 比如感染性疾病以及癌症。 lose their lives due to liver and oral cancer. 每年,有數千名病人 Our best way to help these patients 因為肝癌或口腔癌而喪命。 is to perform early detection and diagnoses of these diseases. 若要幫助這些病人的最好方法 So how do we detect these diseases today, and can artificial intelligence help? 就是早期偵測並診斷出這些疾病。 In patients who, unfortunately, are suspected of these diseases, 現今我們要如何偵測出這些疾病? 人工智慧能幫得上忙嗎? an expert physician first orders 對於很不幸被懷疑可能 得了這些疾病的病人, very expensive medical imaging technologies 專業的醫生首先會囑咐 such as fluorescent imaging, CTs, MRIs, to be performed. 採用非常昂貴的醫療成像技術, Once those images are collected, 例如螢光成像、 電腦斷層掃瞄、核磁共振。 another expert physician then diagnoses those images and talks to the patient. 一旦收集到了這些影像, As you can see, this is a very resource-intensive process, 會有另一位專業醫生根據 這些影像做診斷,並和病人談。 requiring both expert physicians, expensive medical imaging technologies, 不難看出,這是非常耗資源的過程, and is not considered practical for the developing world. 需要專業的醫生 和昂貴的醫療成像技術兩者, And in fact, in many industrialized nations, as well. 而這在開發中國家是不實際的; So, can we solve this problem using artificial intelligence? 事實上,在許多工業化的國家亦然。 Today, if I were to use traditional artificial intelligence architectures 所以,我們能用人工智慧 來解決這個問題嗎? to solve this problem, 現今,若我要用傳統人工智慧架構 I would require 10,000 -- 來解決這個問題, I repeat, on an order of 10,000 of these very expensive medical images 我會需要一萬—— first to be generated. 我重覆一次,大約一萬張 這種非常昂貴的醫療影像 After that, I would then go to an expert physician, 先被產生出來。 who would then analyze those images for me. 產生出來後,接著去找專業醫生, And using those two pieces of information, 來為我分析這些影像。 I can train a standard deep neural network or a deep learning network 用這兩種資訊, to provide patient's diagnosis. 我就能訓練標準的 深度類神經網路或深度學習網路 Similar to the first approach, 來提供對病人的診斷。 traditional artificial intelligence approaches 和第一個方法很類似, suffer from the same problem. 傳統人工智慧方法 Large amounts of data, expert physicians and expert medical imaging technologies. 也會遇到同樣的問題。 So, can we invent more scalable, effective 大量的資料、專業醫生, 以及專業醫療成像技術。 and more valuable artificial intelligence architectures 我們是否能發明 更有擴展性、更有效, to solve these very important problems facing us today? 且更有價值的人工智慧架構, And this is exactly what my group at MIT Media Lab does. 來解決我們現今所面臨的 這些非常重要的問題? We have invented a variety of unorthodox AI architectures 這就是我的團隊在麻省理工學院 媒體實驗室在做的事。 to solve some of the most important challenges facing us today 我們已經發明了多種 非正統的人工智慧架構 in medical imaging and clinical trials. 來解決我們現今在醫療成像 及臨床實驗方面 In the example I shared with you today, we had two goals. 所面臨的一些最重要的挑戰。 Our first goal was to reduce the number of images 在今天我和各位分享的 例子中,我們有兩個目標。 required to train artificial intelligence algorithms. 我們的第一個目標是要減少 Our second goal -- we were more ambitious, 訓練人工智慧演算法 所需要的影像數量。 we wanted to reduce the use of expensive medical imaging technologies 我們的第二個目標—— 我們的野心更大, to screen patients. 我們想要減少使用昂貴醫療成像技術 So how did we do it? 來篩選病人。 For our first goal, 我們要怎麼做? instead of starting with tens and thousands 針對第一個目標, of these very expensive medical images, like traditional AI, 不像傳統人工智慧一開始 we started with a single medical image. 要用到數萬張非常 昂貴的醫療影像, From this image, my team and I figured out a very clever way 我們反而從單一張醫療影像開始。 to extract billions of information packets. 從這張影像,我和我的團隊 想出了一個非常聰明的方法 These information packets included colors, pixels, geometry 來取出數十億個資訊封包。 and rendering of the disease on the medical image. 這些資訊封包包括用 顏色、像素、幾何學, In a sense, we converted one image into billions of training data points, 在醫療影像上呈現疾病。 massively reducing the amount of data needed for training. 在某種意義上,我們是把一張影像 轉變為數十億個訓練資料點, For our second goal, 大大減少了訓練所需要的資料量。 to reduce the use of expensive medical imaging technologies to screen patients, 至於第二個目標, we started with a standard, white light photograph, 也就是減少使用昂貴的 醫療成像技術來篩選病人, acquired either from a DSLR camera or a mobile phone, for the patient. 我們一開始使用的是 一張病人的標準白光照片, Then remember those billions of information packets? 可以用數位單眼相機或手機來拍攝。 We overlaid those from the medical image onto this image, 接著,還記得 那數十億個資訊封包嗎? creating something that we call a composite image. 我們將那些來自醫療影像的 封包疊到這張影像上, Much to our surprise, we only required 50 -- 創造出我們所謂的合成影像。 I repeat, only 50 -- 很讓我們驚訝的是, 我們只需要五十張—— of these composite images to train our algorithms to high efficiencies. 我重覆一次,只要五十張—— To summarize our approach, 這種合成影像,就能把我們的 演算法訓練到很高效能的程度。 instead of using 10,000 very expensive medical images, 總結一下我們的方法, we can now train the AI algorithms in an unorthodox way, 我們不需要使用一萬張 非常昂貴的醫療影像, using only 50 of these high-resolution, but standard photographs, 我們現在可以用非正統的方法 來訓練人工智慧演算法, acquired from DSLR cameras and mobile phones, 只要用五十張高解析度的 一般標準照片, and provide diagnosis. 用數位單眼相機或手機來拍攝即可, More importantly, 這樣就能提供出診斷結果。 our algorithms can accept, in the future and even right now, 更重要的是, some very simple, white light photographs from the patient, 在未來,甚至在現在, 我們的演算法能接受 instead of expensive medical imaging technologies. 病人非常簡單的白光照片, I believe that we are poised to enter an era 取代昂貴的醫療成像技術。 where artificial intelligence 我相信我們已經準備好 要進入一個新時代, is going to make an incredible impact on our future. 在這個時代,人工智慧 And I think that thinking about traditional AI, 將會對我們的未來有很大的衝擊。 which is data-rich but application-poor, 想想傳統人工智慧, we should also continue thinking 它在資料上很豐富, 但在應用上很有限, about unorthodox artificial intelligence architectures, 我們應該要持續思考 which can accept small amounts of data 有沒有其他非正統的 人工智慧架構, and solve some of the most important problems facing us today, 能夠接受更少量的資料, especially in health care. 並解決一些現今我們 面臨最重要的問題, Thank you very much. 特別是健康照護問題。 (Applause) 非常謝謝。
B1 中級 中文 美國腔 TED 影像 人工 醫療 昂貴 演算法 【TED】Pratik Shah:AI如何讓疾病診斷變得更容易(How AI is making it easier to diagnose disease | Pratik Shah)。 (【TED】Pratik Shah: How AI is making it easier to diagnose disease (How AI is making it easier to diagnose disease | Pratik Shah)) 4527 113 林宜悉 發佈於 2021 年 01 月 14 日 更多分享 分享 收藏 回報 影片單字