台灣人工智慧學校南部智慧醫療專班開學典禮 - 主題演講：邁向智慧醫療新時代（陳昇瑋執行長）
?
4 likes?493 views
This document discusses various topics related to artificial intelligence and machine learning. It provides examples of how deep learning is being used for tasks like detecting diabetic eye disease, classifying arrhythmias from ECG signals, and localizing tumors in medical images. The document also notes limitations of current AI, such as its lack of common sense, and discusses how machine learning is being applied in other domains like predicting hospital readmissions, personalized medicine, and monitoring rainforests for illegal logging.
![0.95
F-score
Algorithm Ophthalmologist
(median)
0.91
“The study by Gulshan and colleagues truly
represents the brave new world in
medicine.”
“Google just published this paper in JAMA
(impact factor 44.405) [...] It actually lives
up to the hype.”
Dr. Andrew Beam, Dr. Isaac
Kohane Harvard Medical School
Dr. Luke Oakden-Rayner
University of Adelaide](https://image.slidesharecdn.com/talk190817-190817063942/85/-38-320.jpg)
台灣人工智慧學校南部智慧醫療專班開學典禮 - 主題演講：邁向智慧醫療新時代（陳昇瑋執行長）
- 3. 10
- 4. 11
- 5. 16
- 9. Keyword in AI Research 21
- 10. 23
- 11. 24 Training a prediction machine by showing examples instead of programming it. -Yann LeCun (prediction machine: /- )
- 12. -/ 25 Find the common patterns from the left waveforms It seems impossible to write a program for speech recognition You quickly get lost in the exceptions and special cases. (狠狠撸 Credit: Hung-Yi Lee)
- 13. A - You said “ ” /I I (狠狠撸 Credit: Hung-Yi Lee)
- 14. -/ 28
- 15. 32
- 16. 33
- 17. 34
- 18. 35
- 19. 36
- 20. 37
- 21. 38
- 22. AI 40
- 23. 41
- 24. - 42
- 25. - 43
- 26. - 44
- 27. - 45
- 28. Deep Learning, Machine Learning, and AI 47
- 29. Healthy Diseased Hemorrhages No DR Mild DR Moderate DR Severe DR Proliferative DR 1 2 3 4 5
- 30. 51 Classical Machine Learning Deep Learning Rule-based System Rule extraction LessDomainExperts, MoreEngineers&MoreAccuracy
- 31. 53
- 33. There is no free lunch 55 Classical Machine Learning Deep Learning
- 37. 71
- 38. 0.95 F-score Algorithm Ophthalmologist (median) 0.91 “The study by Gulshan and colleagues truly represents the brave new world in medicine.” “Google just published this paper in JAMA (impact factor 44.405) [...] It actually lives up to the hype.” Dr. Andrew Beam, Dr. Isaac Kohane Harvard Medical School Dr. Luke Oakden-Rayner University of Adelaide
- 39. Deep Learning for Detection of Diabetic Eye Disease 73 Algorithm’s F1-score: 0.95 Median F1-score of 8 ophthalmologists : 0.91
- 40. 74
- 41. AI 75 during 2008 – 2013 in New York City , , , etc. AI , 63% , 5% AI vs. Judges , (!) , / ( , 1000 ) 75
- 43. 77
- 44. 78
- 46. 81
- 47. arxiv.org/abs/1703.02442 Tumor localization score (FROC): model: 0.89 pathologist: 0.73 (狠狠撸 Credit: Google Brain)
- 48. 88 Deep Learning for Kidney Function Classification and Prediction using Ultrasound-based Imaging Chin-Chi Kuo1, Chun-Min Chang2, Kuan-Ting Liu2, Wei-Kai Lin2, Chih-Wei Chung1, and Kuan-Ta Chen2 1 Big Data Center, China Medical University Hospital, China Medical University, Taichung, Taiwan 2Institute of Information Science, Academia Sinica, Taiwan eGFR ( )
- 49. 89
- 50. 10 XGBoost Models Ensemble Accuracy 0.851 Precision 0.870 Recall 0.667 F1 score 0.757 AUC 0.91
- 51. Cardiologist-Level Arrhythmia Detection with Convolutional Neural Networks 95 Goal: diagnose irregular heart rhythms, also known as arrhythmias, from single-lead ECG signals better than a cardiologist
- 52. Input and Output Input: a time-series of raw ECG signal The 30 second long ECG signal is sampled at 200 Hz From 29,163 patients Output: a sequence of rhythm classes The model outputs a new prediction once every second Total 14 rhythm classes are identified 96
- 53. 97
- 54. Model 34 layers NN 16 residual blocks 2 conv layers per block Filter length = 16 samples # filter = 64*k, k start from 1 and is incremented every 4-th residual block Every residual block subsamples its input by a factor of 2 98
- 55. Results – F1 score 99
- 56. 101
- 57. 102
- 58. 103 * 28 out of 142 patients were labeled as depressed.https://www.csail.mit.edu/news/model-can-more-naturally-detect-depression-conversations
- 59. Synthetic Speech Generated from Brain Recordings 104https://www.ucsf.edu/news/2019/04/414296/synthetic-speech-generated-brain-recordings
- 60. Synthetic Speech Generated from Brain Recordings 105 https://www.sciencemag.org/news/2019/01/artificial-intelligence-turns-brain-activity-speech
- 62. Predictive tasks for healthcare Given a large corpus of training data of de-identified medical records, can we predict interesting aspects of the future for a patient not in the training set? ● will patient be readmitted to hospital in next N days? ● what is the likely length of hospital stay for patient checking in? ● what are the most likely diagnoses for the patient right now? and why? ● what medications should a doctor consider prescribing? ● what tests should be considered for this patient? ● which patients are at highest risk for X in next month? Collaborating with several healthcare organizations, including UCSF, Stanford, and Univ. of Chicago. Have early promising results.
- 63. 108
- 64. DeepVariant: Creating a universal SNP and small indel variant caller with deep neural networks 110
- 65. 111
- 69. 117 Rajkomar, Alvin, Jeffrey Dean, and Isaac Kohane. "Machine learning in medicine." New England Journal of Medicine380.14 (2019): 1347-1358.
- 70. The 2014 HSBC Expat Explorer survey rates healthcare 118 https://expathealth.org/healthcare/top-5-health-care-expats/
- 72. (Moravec’s Paradox) High cognitive processes Conscious processes Chesses, math, problem solving Difficult for humans Easy for computers Low Cognitive processes Perception, action, fight/flight responses, social interactions Easy for humans Difficult for computers 120
- 73. 125
- 74. 126
- 76. 130
- 77. Strong AI Weak AI Can think Own conscious Act as it can think Consciousless (1980)
- 78. What we can and cannot today What we can have Safer car, autonomous car Better medical image analysis Personalized medicine Adequate language translation Useful but stupid chatbots Information search, retrieval, filtering Numerous applications in energy, finance, manufacturing, commerce, law, … What we cannot have (yet) Machine with common sense Intelligent personal assistants “Smart” chatbots Household robots Agile and dexterous robots Artificial General Intelligence (AGI) 132 (Credit:Yann LeCun)
- 79. 134 2015 207 31,000 3,700 90% Markram
- 80. 139
- 81. 141
- 82. Generating adversarial patches against YOLOv2 142 https://www.youtube.com/watch?feature=youtu.be &v=MIbFvK2S9g8&app=desktop
- 83. AI Don’t Know What They are Talking About 144 https://www.facebook.com/playgroundenglish/videos/629372370729430/?hc_ref=ARQ HCaS2GZ9jUgZermEupF5yerADq2X9F9P40OR3n70poUiCy7R0X3oHrGxyLSrWVdI
- 84. Change is the only constant. - Heraclitus (535 BC - 475 BC) (狠狠撸 Credit: Albert Chen)
- 86. 154