2. What is AI (or should be better to talk about Big Data and ML)?
The Good part
Mathematics and algorithms are evolving
Multiplex measurements drastically changed the way to deal with data
Resources are growing at a high-speed rate (except for the computation power)
A multi-disciplinary approach is giving fascinating outcomes
Big Data not (only) in term of size but also its purpose
3. What is AI (or should be better to talk about Big Data and ML)?
The bad part
AI/ML still too young and general (supervised, unsupervised, genetic algo, ...)
No, deep learning is not an answer for everything
Cleaning (remove noise, labelling, …) & managing data
… are still one of the most important, although underestimated, steps (need a different approach)
4. What is AI (or should be better to talk about Big Data and ML)?
The ugly part
We need more data and more OPEN(less business on the patient’s data)
Research needs to be more inclusive and OPEN
Privacy and open data: an endless war?
Regulations (!!!)
PS: Did I already say OPEN ?
5. Where the big data from medical sectors are pushing research and trials
Push domain expert (doctors and researchers) to join tech and math/data people (data scientists +
computer engineers)
Need to use different approach together (supervised and unsupervised learning)
Improving data collection, storage, cleaning, processing and interpretation continue to develop, although
not always by, or for, medical researchers.
Experimental approach: hypothesis-free to the rescue
(labelling and matching unknown part of the system => AI)
6. How AI makes this easy. Why is the AI wrong sometimes? How to avoid it
How to? 2 words: Big Data!
Big data refers to the vast quantities of information created by the digitization of everything, that gets
consolidated and analyzed by specific technologies.
Applied to healthcare, it uses specific health data of a population (or of a particular individual) and
potentially help to prevent epidemics, cure disease, cut down costs, etc.
Machine learning and supervised learning could help understanding, labelling, cleaning and organizing a
vast amount of data (algo like KNN/K-Means, Regression, SVN/SVV, …)
Unsupervised learning, the (huge) family of deep learning algos (CNN, GAN, …) can help to find more
and different pattern inside the same data (sometimes less dependant on previous/domain knowledge)
7. How AI makes this easy. Why is the AI wrong sometimes? How to avoid it
What could possibly go wrong?
Missing “data collection” culture
Have you ever heard of “Doctor Google”?
Without domain expert (doctors, researchers, …) the simple collection and analysis of data could be,
sometimes, meaningless or even dangerous
Regulations are quite behind and could become a problem as a late joiner.
Storage and privacy are quite “interesting” challenges.
8. Old gathered data vs new gathered data: the changes
Where are we?
We are missing a proper set and amount of data for an individual
Healthcare is behind on collecting and managing data
Infrastructure is mostly dated and not well suited for sharing
Medical data are carefully guarded and pretty private (not to mention quite complex)
9. Old gathered data vs new gathered data: the changes
What can we do?
Improve the structure allowing to gather, manage and SHARE data accurately
Anonymous data and protected data (encrypted, advanced cryptography, compressed, …)
Clean data: domain expert join data expert to provide a better and more focused process
10. Old gathered data vs new gathered data: the changes
Take away
Need for standardization of data content, format, and clinical definitions, collaborative networks with
sharing of both data and expertise (and/or research results/papers)
Teasing point
Do we need to reconsider how and when the analytic methodology is taught to medical researchers?