Neurology and Data Science

Last week I had a chance to visit USC's LONI: Laboratory of Neuro Imaging. Thanks to Farshid Sepehrband (@fsepehrband) for the invite and tour! The best part for me about doing data skeptic is when I have opportunities to visit fantastic companies and labs doing really interesting work.

I am frequently surprised how nine overlapping the machine learning and artificial intelligence worlds are with the discipline of neurology. I think I have a high level understanding of why this is, but that's the post for another day.

Broadly speaking, LONI six to improve understanding of the brain in health and disease. The manner in which they're doing this is fairly multifaceted. Naturally, the lab has on MRI machine. In fact they have a more powerful one about to come online as well. In addition to their own research, I believe outside groups are able to gain access to this facility as well. The MRI machines generate an enormous volume of data with each scan.

There's a push to decrease the resolution of these devices. I was deeply impressed to hear that the lab was achieving submillimeter resolution of the images they are able to take of the brain. They shared with me some imaging done on older technology (or at least down sampled to share the idea) compared to their most advanced scans today. It was a dramatic difference. About as dramatic is the clarity we had for the planet Pluto pre-and post-new horizons. To me, it's almost a miracle the field had made any advancements before the higher resolution images, as it was startling to see how much new detail was arising in there best scans.

Every scan generates an enormous amount of data. Luckily, the lab is also equipped with an on premises, Fairly substantial data center. It's a striking visual right on the first floor as you approach the building that houses this team. Diagnostics on the data center wall share insights about how the computer cluster is being put to work. This data center is not only a great tool for the staff of LONI lab, but also enables the work of other groups that (presumably) don't have the same facilities. LONI is building the infrastructure to handle the massive size of MRI imaging corpora and make it easier for groups to share data and collaborate.

The demands of this massive data set has also require the lab to employ a wide variety of software developers, data visualization people, and data engineers. They've developed a number of software packages that are available to help with various needs a lab like there's would have. The LONI Pipeline in particular is very interesting to me. It resembles a number of ETL tools I've seen. Yeah it also has the hallmarks of an intuitive interface enabling non-technical users to describe the flow of their data, the operations to perform on it, and the persistent steps to take when the work is done.

The lab is making use of machine learning in some of their analysis. We are planning to put together an episode on the lab in the near future, so I'll leave so that discussion for your future ears.

My outsider impression of neurology has long been that it's a field in its infancy. Perhaps some would agree. The MRI is a tool of measurement is sometimes compared to a Satellite Image of the earth. Often very useful! Yet very low resolution for seeing details on the scale that human beings typically live. Although I imagine there is still a long ways to go, the resolution of cutting-edge machines is pretty remarkable, and is ready to provide valuable insights to the medical world today.

Neurological disorders such as Alzheimer's effect this also seems to be a growing category of fatality, as formerly life threatening conditions slowly migrate into the treatable and curable realms. I was shocked to see how are obviously different the brains of a healthy person and an Alzheimer's patient were. Perhaps the patient was in an advanced stage, but the images I saw were so obviously different gnome machine learning was required. This doesn't apply strongly that the measurements they're collecting coupled with Advance machine learning approaches, are likely to be an incredibly useful for early detection and localization.

Overall this is a really inspiring as it. I'm so glad I was able to go. While the data science world might not be learning much from neurology in the sense that what we call neural networks have little or nothing to do with the brain, in the field of AI arguably hasn't been able to ask many useful questions for neurologists to answer, I was delighted to learn how much data science is contributing to these worthwhile pursuits. We may be a long ways away from having a deep understanding of the workings of the brain, but I it's inspiring to see labs like this that are helping to improve on our understanding.

Stay tuned today to skeptic. While the production might take us a little while, I think you'll be happy with the episode we're planning.