This post was originally published on this siteI have blogged a decent amount recently about VVols and in many of those posts I mention config VVols. When using ...
Over a couple beers with a friend and colleague, Par Botes, we were excitedly talking about Artificial Intelligence (AI), Machine Learning, and Business Intelligence (BI) in healthcare. Specifically, our conversation got very interesting when I told him about radiomics (read about it in one of my past blogs here). He paused me and started giving me a discourse on S/N (Signal-to-Noise Ratio) because, well, that’s Par. However, he’s bang-on — the similarities between S/N in electricity and healthcare are manifold.
The goal for S/N in electricity is to maintain a positive ratio such that the signal is louder and clearer than the noise. A significant amount of engineering goes into the physical world to improve the S/N so that we can experience more accurate measurements, better phone calls, optical transmissions, and so on and so forth. Also, by no means am I the first to write about S/N in healthcare — for example, read Kim Bellard’s take on this here.
Now, let’s get into the meat of my thoughts by reviewing radiomics; radiomics is defined as “the conversion of images to higher-dimensional data and the subsequent mining of these data for improved decision support.”1 Yes, improved decision support! Yes, mining! Yes, higher-dimensional data!
Radiomics started out focused on oncology studies, but is slowly starting to gain acceptance and adoption (and clinical diagnosis approval by the FDA) across other modalities. In the most basic form, radiomics can be thought of thusly: let’s convert images into data, then let’s apply algorithms against that data to identify potential areas of interest from the images. These algorithms are on the basis of known patterns — that is, we instruct machines to search for these known disease / abnormal patterns and report back any potential matches so that we can investigate. This has huge impact on individual as well as population health, where the “… analysis promises to increase precision in diagnosis, assessment of prognosis, and prediction of therapy response.”2 Wonderful!
Let’s take radiomics to another level. What if, in addition to searching for known patterns, we unleashed Machine Learning and AI on this radiomics data, with the goal of identifying net-new patterns across patient cohorts (for example, patients who suffer from a particular chronic condition)? What if we then applied more computer science to take these newly discovered patterns and overlay them on top of radiomics data for our entire healthy patient population? Could we use this to predict whether any of our currently healthy patients share patterns with our known sick patients? Could we then use this analysis to predict predisposition to diseases / conditions? Sounds a lot like what we’re trying to accomplish with genomics, only done with imaging data that’s available in droves, and is relatively “easy” to mine and analyze.
Improving speed-to-diagnosis and intervention by applying Machine Learning and AI to voluminous imaging data is most definitely improving the S/N in healthcare. To do this right, however, you need the right infrastructure (hence the “” around easy up above).
In a subsequent post I’ll talk about how we, Pure Storage, are helping revolutionize healthcare IT infrastructure, and making it disappear into the background. Till then, let’s band together to dream up additional ways in which we can apply computer science and mathematics to accomplishing better patient care and outcomes!
1 RSNA. (2016). Radiomics: Images Are More Than Pictures, They Are Data. Retrieved from: http://pubs.rsna.org/doi/full/10.1148/radiol.2015151169
2 RSNA. (2016). Radiomics: Images Are More Than Pictures, They Are Data. Retrieved from: http://pubs.rsna.org/doi/full/10.1148/radiol.2015151169