AI / MACHINE LEARNING IS MAINSTREAM NOW — RSNA(I) 2018 IN AN EXPONENTIAL WORLD

The title of my 2016 report was: BEYOND IMAGING: Radiology is Changing! Big Data / Machine and Deep Learning / Artificial Intelligence is everywhere!“ And the title of the 2017 report: Machine Learning in the center of Radiology Innovation

What was a beginning, followed by careful acceptance has now become THE major and most often referred innovation at pretty much any booth ... within just 3 years! The hardware innovations are becoming the enablers for the post-processing applications. We are living in an exponentially developing world ... and therefore this years header title is: AI / MACHINE LEARNING IS MAINSTREAM NOW — RSNA(I) 2018 IN AN EXPONENTIAL WORLD (the “RSNAi” with permission ?stolen“ from a report of Mathias Goyen).

Before I start my write-up of the latest and greatest developments in radiology from the annual RSNA conference in Chicago allow me to report some societal developments that the Washington Post / NYT reader and occasional FOX watcher observed talking to our american friends and partners ... which worries and scares me. The US has become a deeply divided country over political direction and leadership. I certainly do have significantly more friends that are politically blue, but there were never really big discussions if you met and talked to a red one. Has all changed now! Heated and almost violent arguments are exchanged and I am shocked about the negative effects of the changed political direction under questionable leadership. Can we get reasonable and less celf-centered again?

Weather: typical november for Chicago with slushy rain, snow and cold temperatures. Attendance: not sure ... my impression says less, but I also heard selected positive comments. The poster hall was certainly not busy ... at least not during the times I was there. Machine Learning Area: bigger than last year with a large number of very innovative

companies and projects. Google was also there for the second year ... interested to see what their next steps in the healthcare segment will be.

Booth Designs and General Impression

GE next to PHILIPS in the North Hall. Siemens Healthineers and Canon in the South Hall and adjacent to Siemens the new big boy from China — United Imaging. This company showed a full line of diagnostic imaging products that even included a linear accelerator and a dynamic full-body PET system. Lots of questions regarding their technology acquisition, the way they deal with IP issues, and the background financing. I am sure that we will see some sideline battles between the established OEM’s and them in the near future.

I generally liked the booth designs and setups of the 4 large providers. Siemens had a two floor open concept with the big systems towards the aisles and the workstations in the middle (I personally believe the other way around would be better) ... and they had an INNOVATION Lounge — access with NDA. So I cannot report what I saw, but if I could it would not be extremely surprising to the “expert”. The Philips booth again quite colorful with only few large systems openly placed. Canon (former “Toshiba” - maybe I have it down by next year) single floor again in striking red and finally GE with inside and outside screens completely circumventing the booth on a second floor height. And they used warm wood appliances. Difficult decision for the in-official MF booth competition ... after consultation with me and myself I will award the first price this year to Philips very closely followed by Siemens and GE (tied). As long as Canon has this extremely thick carpet they will never win!

Just a couple of short notes on other vendors: Samsung greatly reduced presence (and also no technical innovations as far as I could tell ... but I could not find someone who was able and willing to show me around) and Carestream even more “orange”. And again, United Imaging with a system packed large booth, but with little traffic ... maybe the attendees are still skeptical ... we will observe on whether that will change next year.

Technological Innovations

I am incapable of selecting and completely judging the AI / ML innovations based on a short introduction. There is perceived nothing working anymore - from US to MRI - without a GPU and a dedicated neural network ... for just about any application you can imagine — workflow improvements, patient data evaluation, dosis reduction, automatic parameter selection, remote analysis, image enhancements, noise reduction, clinical reporting, image feature extraction, automatic evaluation and warning ... and many more. So one more time ... AI / ML / Deep Learning is definitely here now, is showing huge improvements and will change the entire radiology workflow and the current way radiology is performed. This is not negative, but unavoidable and it is clear that where an artificial system can replace a human it will and probably should. The AI in the coming years will lead to a cooperation between humans and smart programs and I am convinced that the healthcare provision will benefit. With that the radiologist will have an important, maybe even more important, role in the near future.

Every year I ask the vendors to show/explain me their personal highlight(s). I typically tell them that I am interested in the near future disruptions or unique offerings rather than to show me the incrementals that are expected. Yes, there were many new things, but the report wants to only focus on a few that I really liked (it is my report after all!). GE introduced a local AI for a critical care radiography system — that is obviously centrally updated with the learnings from all the other systems — that can detect a pneumothorax once the image has been taken. AI that most certainly can save lives. And, I also liked their improved AIR MRI coils in very open and light designs.

Siemens impressed me with their DIGITAL TWIN concept that combines all the patients data files (genetic, lab, imaging, ...) for future referencing, evaluation, analysis and with that for prevention. Certainly an obvious thing to do and develop, but looks like it is executable already. 

The Philips 7 liter Helium MRI magnet without quench-pipe is cool and will most likely also be a desired innovation with the other vendors. I was playing around with the AZURION cathlab software suite that includes advanced image combinations and catheter options — very impressive.

And finally Canon with an AI program that is capable to selectively reduce the image noise and with that is able to improve the Signal to Noise ration dramatically (or reduce imaging time accordingly). 

I was a little critical on the MR/LINAC systems last year — interestingly enough ... no or no obvious mention anywhere. What happened?

Lots of other little innovations — even without AI — for dedicated Breast CT or PET. These are not completely new and I am trying to find out on whether there really is a clinical need and an economic benefit. I am not sure!

Also goes for neonatal MRI — seen three companies now that have introduced dedicated high-field systems, but I am not certain yet that these systems will actually find a large number of buyers due for their limited application. But I am always happy to see new devices and actually hope that all will be successful. I really like however when companies come up with products that actually make a clinical process cheaper rather than always introducing new expensive and highly selective systems. These incremental innovations typically make healthcare more complicated, more expensive and less democratic.

Having said this United Imaging showed their Explorer PET - a joint development with the UC Davis - that is able to present dynamic whole-body studies. I have not found out what the pricing is, but rumors say that it is in the higher single digit millions.

On the other hand there was a small company (KOIOS) in the Machine Learning section that showed an algorithm that evaluates and analyses thousand of features on ultrasound images and can then provide a confidence map on whether a lesion is benign or not. Ultrasound is growing again!

That also brings me to the end ... it is very obvious that radiology / imaging will be changing dramatically in the coming years with further developments in AI / ML / DL, sensors, digital biology, synthetic biology, 3D printing and other exponential technologies. Please think back just 10 years or 5 years and remember what the highlights at RSNA were then! Significant changes ... and be honest with yourself ... you did not foresee these developments in such a short period of time. Well, project forward! ... and think exponential!

This year we (my university chair and associated institutions and companies) had our own booth (S1239/1240) under the header GERMAN INTERVENTIONAL MRI community. We were very happy with the attendance and interest in our prototype designs. And, we are also happy that we (one of my PhD students) were one of 5 awardees of the GE E10 liver grant that was announced during the RSNA. 

Look forward meeting you next year!

An exciting 12 month to everyone!