Abstract submitted to Vitamin D Workshop, New York, May 2019
W.B. Grant 1, H. Lahore 2 , 1 Sunlight, Nutrition, and Health Research Center, San Francisco, CA, U.S.A.; 2 VitaminDWiki, Port Townsend, WA
Artificial Intelligence/Machine Learning (AI) is any computer program that uncovers relationships in vast amounts of data. Often these associations are previously unknown or unsuspected. AI is maturing and is already beginning to impact all fields of study, including research and medical practice. As computing power increases the cost goes down, and the quantity and quality of genetic and health data of individuals increases exponentially, the opportunities for health professionals to exploit the data with AI become enormous.
In just the past 3 years there has been a 1,000 times increase in the computing power which can be applied to an AI task. So the same task can be completed 1000X faster, or the additional compute power can provide 1000X deeper analysis in the same amount of time.
Using AI, Vitamin D researchers will within the next 5 years be able to improve the design of their experiments based on analysis of data in studies including data from previous experiments, set optimum dosages, and take into account multiple real factors that will affect the success of outcomes.
Using raw data from individual studies rather than relying solely on meta-analyses will improve the confidence of RCTs.
Researchers will be able to use data from “failed” experiments, get proofs from wide-spread but incomplete data, and get preliminary results very quickly. Observational non-RCTs, which make up a majority of published research, can often be modified in real time, based on instant analysis of ongoing data.
Medical practitioners will be able to tailor their recommendations to individuals based on data available from hundreds to millions of similar cases, and can be quickly and comprehensively informed of updates on specific diseases.
Truly, the future of AI in Vitamin D research and practice is mind-boggling.
If this abstract is accepted, this page will have the entire presentation in June, 2019
Deep Medicine, by Eric Topol - March 2019
Excellent review of 100+ wide-ranging applications of AI in Medicine
Already applications are developed and underway - costing $1 million to $600 million
I continue to estimate the AI applications to Vitamin D to cost < $1 million
Some notes from the book
30% of Breast Cancer surgeries could be avoided by AI analysis of images
Book has a nice concept: AI finds your "digital Twin" in the mountain of data
AI is doing speech analysis for 30+ speech parameter to predict Parkinson's and many other health problems
AI analysis of images (Xray, CT) can be done with 1/10 to 1/100 as much energy
thus CT scans could be done with AI in < 10 minutes
AI is moving much faster in drug discovery and research than in areas which are slowed up by the FDA
All health-related things need to be individualized - due to gut. gene, etc
Reactions differ between Individuals: food eaten (e.g. bread) , drugs, supplements (vitamin D), etc
I disagree with Topol - Radiologists and pathologists should fear for their jobs
AI can already find microfractures as small as 0.01% of the image
Metastasis detection: AI 92%, Pathologists 73%
I would hope that AI could slow up the rising health care costs in < 5 years
Table of Contents
chapter one INTRODUCTION TO DEEP MEDICINE
chapter two SHALLOW MEDICINE
chapter three MEDICAL DIAGNOSIS
chapter four THE SKINNY ON DEEP LEARNING
chapter five DEEP LIABILITIES
chapter six DOCTORS AND PATTERNS
chapter seven CLINICIANS WITHOUT PATTERNS
chapter eight MENTAL HEALTH
chapter nine AI AND HEALTH SYSTEMS
chapter ten DEEP DISCOVERY
chapter eleven DEEP DIET
chapter twelve THE VIRTUAL MEDICAL ASSISTANT
chapter thirteen DEEP EMPATHY
- The Pediatric AI That Outperformed Junior Doctors Singularity Hub Feb 2019
AI was trained on 567,498 patients with some 101.6 million data points,55 different diagnosis codes
"Essentially, it took nearly 15 years before a physician could consistently out-diagnose the machine"
- Excellent prediction of very low vitamin D in elderly from just 16 questions (analyzed by ML) – June 2017
- Vitamin D estimation nicely improved by neural networks – May 2015
- Model has 80 percent chance of predicting vitamin D levels to within 10 ng – Feb 2012
- Predict Vitamin D category listing has
54 items along with related searches
- Toward predicting vitamin D levels without a blood test. thoughts on the subject
Second abstract submitted VitaminDWiki: your go-to Vitamin D website - 2019
Short URL= is.gd/AIVitD