Recommended
More Related Content
What's hot
What's hot (19)
Similar to Queen's Grand Rounds - Artificial Intelligence at ASTRO 2019 (Nov 14 2019)
Similar to Queen's Grand Rounds - Artificial Intelligence at ASTRO 2019 (Nov 14 2019) (20)
Recently uploaded
Recently uploaded (20)
Queen's Grand Rounds - Artificial Intelligence at ASTRO 2019 (Nov 14 2019)
- 1. Artificial Intelligence in Medicine ASTRO 2019 Review November 14, 2019 Ian Pereira
- 2. Outline Overview of AI at ASTRO 2019 Introduction to AI Modern AI Approaches for Oncology Ethical and Explainable AI Relevance to Clinical Oncology Research & Practice Considerations for Supporting AI in Institutions
- 3. Objectives REVIEW THE AI PROJECTS AT ASTRO 2019 UNDERSTAND THE BASICS OF AI FOR ONCOLOGY RECOGNIZE WHAT APPROACHES WORK APPLY FINDINGS TO RESEARCH AND PRACTICE
- 4. Artificial Intelligence at ASTRO 2019 – A Greater Presence?
- 5. AI at ASTRO 2019: Ethics, Data Science, Physics, Clinical Practice, & the Future • Ethical Challenges of AI (David Magnus) • AI and Augmented Intelligence: Where do we stand? (Suchi Saria) ASTRO Presidential Symposium: AI and Deep Learning in Medicine • Introduction to Big Data Analytics (Jun Deng) • Machine Learning Algorithms in a Nutshell (Issam El Naqa) • Applications of AI in RO (Mary Feng) Introduction to Big Data Analytics and Artificial Intelligence • How we Treat for Cure in Metastatic Disease (Kristy Brock) • Toward Cure in Metastatic Disease: The Role of AI (David Jaffray) AI Will Drive Who We Treat for Cure With Mets & How We Treat Them
- 6. AI at ASTRO 2019: Ethics, Data Science, Physics, Clinical Practice, & the Future •What Department Chairs need to know for supporting & promoting growth of Big Data (Theodore Lawrence) •NCI’s IT for Cancer Research Funding Opportunities (Bhadrasain Vikram) •What Physicians Need to Know in Building Careers and Teams in Big Data (Clifton Fuller) •Choosing the Bias: A Physicist’s Perspective on Informatics (Dan Ruan) •Emerging Career Paths in RO Big Data (Jayashree Kalpathy-Cramer) Emerging Career Paths in Radiation Oncology Big Data •Awardee: ML to Predict Acute Radiation Toxicities for Head and Neck Cancer Patients (Reddy) Multiple Abstract Presentations •VARIAN ETHOS: AI-driven Adaptive Treatment Planning •RayStation RayCare: Man vs. “Machine” for Treatment Plan Optimization Vendor Releases
- 7. Intro to AI for Oncology
- 8. Medicine is exceeding human cognitive capacity David Jaffray, MDACC
- 9. Yet our understanding is still limited Theodore Lawrence (Ann Arbor)
- 10. This results in limited models, today Theodore Lawrence (Ann Arbor)
- 11. However, future demands are more complex Theodore Lawrence (Ann Arbor)
- 12. Is health informatics the solution?
- 13. An unlikely solution, alone Friedman, 2009
- 14. Friedman, 2009 Together is better
- 15. Why Now?
- 16. Why Now? David Jaffray, MDACC
- 17. Why now? Theodore Lawrence (Ann Arbor)
- 19. Now, the DL of AI is on the rise Cardenas, 2019 Coccia, 2018 Cui, 2019
- 20. Learning approaches are evolving Suchi Sariya Easier Harder We are here
- 21. And improving clinical decision-making in some fields
- 23. Ethical Challenges for AI • New Models of Data Stewardship • Duty to patients, data, & providers • Duty to make good use of data • Responsible third-party access • More Awareness of Biases • Recognize missing data from under- represented populations • Avoid self-fulfilling predictions • Recognize differences in health outcomes can be a function of social injustice • More Trust • Costs of scaling quickly (EHR, alarms, CBME) David Magnus
- 24. Ethical Solutions for AI • Transparency • Culture shift in data stewardship • Close collaboration between HCPs and developers • Awareness of bias through education, diversity & inclusion • Strive to reduce disparities in all algorithms David Magnus
- 25. Opening the Black Box - Interpretability
- 26. Shifting Culture Faden et. al
- 27. So What? Relevance to Clinical Oncology Research & Practice
- 28. AI may improve decision-making, efficiency, quality, & safety at multiple points Weekly Review Dosimetry Contouring Mary Feng
- 29. Decision Support? Objective: To develop predictive models of acute toxicity during radiation for HN cancer patients. •Unplanned hospitalization (<3 months from RT start) •Significant weight loss (>10% during RT) •Feeding tube placement Reddy, 2019
- 33. This study demonstrates the feasibility of employing precision oncology to predict acute radiation toxicities. May facilitate the identification of patients for whom early intervention is warranted. Reddy, 2019
- 35. DL for Autosegmentation? Nikolov, 2018Deep Learning methods have improved OAR and Target Volume Auto-Contouring (over atlas-based segmentation), but requires a large learning library.
- 37. Augmented Adaptive Treatment Delivery?
- 39. • AI is being studied across the radiation oncology workflow • Machine Learning (ML) applications are maturing, now with clinical decision-making being studied. • Deep Learning is the next frontier, with newer learning approaches making gains in auto-segmentation and starting to improve contouring speed, and further improving treatment planning quality and speed. • AI is not yet seeing widespread clinical use, but is an option from commercially available systems (VARIAN RapidPlan, RaySearch, & Mirada’s DLCExpert)
- 41. Foster mentorship and team skills
- 42. Foster mentorship and team skills
- 43. Grow a data culture (don’t just mine) • Clinical staff with “hands on” skills bridging expertise • Data support staff • Mainstream tools • Standardization Chuck Mayo, 2016 Digga, Gold
- 44. Use AI Studies to Augment Traditional Trials Mayo et al. Big Data in Designing Clinical Trials: Opportunities and Challenges. Front Oncol. 2017 Not replace them.
- 45. Consider training As a field, do you think that training provided by radiation oncology in bioinformatics is sufficient? Do you think that training in bioinformatics would allow you to advance your research career?
- 46. Summary • Interest in AI continues to increase • Modern approaches use deep learning to tackle “messy” health data for increasing gains • Frameworks for ethical AI are being developed • AI Applications for clinical oncology are increasing • Support at institutions for trainees, faculty, and the data itself may help ensure best local use
- 47. Thank you.
Editor's Notes
- Emerging Career Paths in Radiation Oncology Big Data - What you need to know. Introduction to Big Data Analytics and Artificial Intelligence
- Emerging Career Paths in Radiation Oncology Big Data - What you need to know. Introduction to Big Data Analytics and Artificial Intelligence
- Jaffray
- Theodore Lawrence (Umichigan)
- Theodore Lawrence (Umichigan)
- Theodore Lawrence (Umichigan)
- Friedman, 2009 (JAMA)
- Friedman, 2009 (JAMA)
- AI: ability of a machine to perform tasks commonly associated with intelligent human behavior. ML: able to appreciate hidden patterns within the data which can then be used to perform a task without explicit programming. Dozens of ML algorithms that have been proposed over the past several decades, and the most traditional forms of ML, such as logistic regression, have proven themselves as valuable tools for general clinical oncology research: DL algorithms are able to learn the optimal features that best fit the data through the training process, avoiding the need to use pre-engineering, unstructured data. This ability has allowed DL algorithms to outperform traditional ML algorithms in many common AI problems, including image classification, natural language processing, and sequence prediction.
- Jaffray
- ML & DL: Cui et al, 2019 (preprint) Auto-Segmentation: Cardena, 2019 Figure 4. Evolutionary trends of number of article about deep learning algorithms per different types of cancer (lung, breast, and thyroid) over 1996-2018 period. https://arxiv.org/ftp/arxiv/papers/1905/1905.06871.pdf
- Suchi Sariya (JHU) Weakly supervised and active approaches may be best for “messy” systems
- 90 clinical units Sariya: Abstract (2015): We analyzed routinely available physiological and laboratory data from intensive care unit patients and developed "TREWScore," a targeted real-time early warning score that predicts which patients will develop septic shock. TREWScore identified patients before the onset of septic shock with an area under the ROC (receiver operating characteristic) curve (AUC) of 0.83 [95% confidence interval (CI), 0.81 to 0.85]. At a specificity of 0.67, TREWScore achieved a sensitivity of 0.85 and identified patients a median of 28.2 [interquartile range (IQR), 10.6 to 94.2] hours before onset. Of those identified, two-thirds were identified before any sepsis-related organ dysfunction. In comparison, the Modified Early Warning Score, which has been used clinically for septic shock prediction, achieved a lower AUC of 0.73 (95% CI, 0.71 to 0.76). A routine screening protocol based on the presence of two of the systemic inflammatory response syndrome criteria, suspicion of infection, and either hypotension or hyperlactatemia achieved a lower sensitivity of 0.74 at a comparable specificity of 0.64.
- Trust
- Mary Feng, UCSD. Patient Assessment: Should I treat? Simulation & Contouring: Will autosegmentation save time? Treatment Planning, QA, & Delivery: “Knowledge-based planning”, predict QA outliers (QA failures, OBI issues, collision risk, machine downtime), adaptation/management (adaptive radiation treatment, PTV under- or OAR over-dosing, need for IV fluids or feeding tubes) Follow-Up: Early or routine, adjuvant treatment?
- Nikolov, 2018
- VARIAN Ethos – for adaptive radiation treatment. Autocontouring. Forecasting. (Similar to RapidPlan?). Tseng, Michigan 2018
- Mentorship to share skills Education time (protected) & resources to build Collaborators Resilience Clinical team, mentor team, look for synergies,
- Mentorship to share skills Education time (protected) & resources to build Collaborators Resilience Clinical team, mentor team, look for synergies,
- Gold digga.
- Mayo, Matuszak, Schipper et al. Big Data in Designing Clinical Trials: Opportunities and Challenges. Front Oncol. 2017; 7: 187.
- Twenty-six department chairs and 91 general respondents submitted responses. Among general respondents, 69% were current trainees and 31% had completed training.