Radiology in the Era of Artificial Intelligence November 11, 2019 - - PowerPoint PPT Presentation
Radiology in the Era of Artificial Intelligence November 11, 2019 Curtis P. Langlotz, MD, PhD Professor of Radiology and Biomedical Informatics, Stanford University Director, Center for Artificial Intelligence in Medicine & Imaging
November 11, 2019
Curtis P. Langlotz, MD, PhD Professor of Radiology and Biomedical Informatics, Stanford University Director, Center for Artificial Intelligence in Medicine & Imaging Associate Chair, Information Systems, Department of Radiology Medical Informatics Director, Stanford Health Care @curtlanglotz
McNitt-Gray MF et al. Computed Tomography Assessment of Response to Therapy: Tumor Volume Change Measurement, Truth Data, and Error. Transl Oncol. 2009 Dec 1; 2(4): 216–222.
Schwartz LH et al. Variability in Response Assessment in Solid Tumors: Effect of Number of Lesions Chosen for Measurement. Clin Cancer Res. 2003 Oct 1;9(12):4318-23.
http://www.image-net.org/
5
Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, et al. ImageNet Large Scale Visual Recognition Challenge. Int J Comput Vis. 2015;115: 211–252. 0% 5% 10% 15% 20% 25% 30% 2010 2011 2012 2013 2014 2015 2016 2017
ImageNet Visual Recognition Error Rates Human error rate
Karpathy, Andrej & Li, Fei Fei. Deep Visual-Semantic Alignments for Generating Image Descriptions, CVPR, 2015 http://www.radiologyassistant.nl/
ASR 33 Teletype, 1976 75 lbs, 110 baud
https://www.pdp8.net/asr33/asr33.shtml
DEC PDP-8, 1965 0.333 MHz, $18,500 Apple A12 Processor, 2019 2.49 GHz x 6, $1000
IBM 5MB HD, 1956 $3200/month lease Seagate 4TB HD, 2019 $99, Amazon.com
Negative Cases Positive Cases
https://hackernoon.com/challenges-in-deep-learning-57bbf6e73bb
A flexible and powerful form of machine learning. Benign Malignant
Faces Cars Elephants Chairs
Courtesy of Andrew Ng
Labeled Training Data New Image Recon Methods
CT scan icon by Sergey Demushkin from the Noun Project
Source “Raw” Data New Image Labeling Methods New Machine Learning Explanation Methods Actionable Advice Decision Support Systems New Machine Learning Methods
Ingredient for Success AI Startups Established Companies Healthcare Delivery Systems Professional Societies Academic Health Systems
Deep technical knowledge High performance computing Interdisciplinary teams Ongoing source of labeled images Infrastructure for prospective evaluation Market dissemination channel =can acquire =available =difficult to acquire
Key C Compute ter S Science F Faculty C Collaborators
Andrew Ng, PhD Deep Learning Pioneer Chris Manning, PhD Stanford AI Lab Director Fei Fei Li, PhD Creator of ImageNet
PE classification Appendicitis Cervical spine fracture 14 chest abnormalities Coronary calcium scoring Fractures Fractures Knee menisci and ligaments Knee cartilage Brain age Brain tumors Renal abnormalities Renal cell carcinoma Deep vein thrombosis Brasfield scoring Bone age Brain abnormalities MR CT US XR Lung cancer Pneumonia Normal/abnormal PT
AI Throughout The Imaging Life Cycle
17
Image protocoling Image recon- struction and enhancement Imaging sequence
Image quality control Image triage Computer aided- detection Disease classification
https://aimi.stanford.edu/
Lower radiation dose and imaging time Answer the clinical question Decrease length of stay Prevent/detect disease early among populations Select precise treatment with global reach Enable patient self- scheduling Reduce repeat imaging
18
Scanner Cohort Management
labeling
Cohort selection
Algorithm Training & Testing
Machine-learning- ready data set
Research Data Warehouse (STARR)
Data Migration
AI Prospective Evaluation and Deployment Platform
UPSTREAM DOWNSTREAM Containerized Algorithm AI results Clinical images
DICOM Router Clinical EMR, RIS, PACS, Reporting
Clinical images Image labels
Imaging Reporting System
AI results AI results Other clinical data Clinical images for AI processing “Raw” detector data for AI processing
Clinical Practice Laboratory
AI Vendor
and validating AI algorithms
integration of AI algorithms
common data elements
to ensure safety and efficacy
https://doi.org/10.1016/j.jacr.2019.04.014
21
https://aimi.stanford.edu/
Measure = 7 ± 6 Measure = 7 ± 2
Adapted from Kevin O’Donnell and Daniel Sullivan
Variability in:
QIBA Profile specifying:
Treat Wait
Engage: @curtlanglotz @StanfordAIMI LinkedIn: aimi-stanford Join the mailing list: https://aimi.stanford.edu/engage/subscribe Apply to become affiliated faculty: https://aimi.stanford.edu/engage/affiliate-faculty