Deep learning and big data in medical imaging: buzzwords and hype? Maybe better focus on workflow and integration.

It is hard to keep up with the latest technologies, let alone, all the new buzzwords or hype introduced by smart marketers to differentiate their products. I believe that it is better to call a product by its name based on what it does. Take “deep learning,” for example, isn’t “deep learning” just another form of neural networks, which is how Wikipedia classifies it, i.e. after a four-paragraph description, it says “Deep learning has been characterized as a buzzword, or a rebranding of neural networks.”

If you follow this quote it leads to another interesting statement from IEEE Fellow Michael I. Jordan, Pehong Chen Distinguished Professor at the University of California, Berkeley:

“The overeager adoption of big data is likely to result in catastrophes of analysis comparable to a national epidemic of collapsing bridges. Hardware designers creating chips based on the human brain are engaged in a faith-based undertaking likely to prove a fool’s errand. Despite recent claims to the contrary, we are no further along with computer vision than we were with physics when Isaac Newton sat under his apple tree.”

After reading these quotes from people who are much smarter than me, I am reminded that maybe we should not get carried away by these new terms and buzzwords and instead stick to what we know and what feels right. If a new development feels like it is hype, it very likely is.

Instead of pursuing a fool’s errand, maybe we should fix what is broken, which is focusing on improving our workflow and integrating our current systems in a better manner. Implementing a VNA, deconstructed PACS, deep learning, applying big data solutions will only make the situation worse if you don’t optimize your current system first.

As an illustration, during our recent webcast on enterprise imaging, we had a poll asking the attendees for the top healthcare imaging and IT priority in their institution, and the top two issues were workflow and integration. These issues are not new as they have come up during past conferences multiple times but they still do not seem to be sufficiently resolved and addressed after all these years.

As a matter of fact, talking with other industry consultants, we seem to agree that more than 50 percent of the current healthcare imaging and IT systems could definitely use a “tune-up.” Think about a V6 in a car running only on four cylinders.  There are plenty of consultants that are busy with HIPAA audits, which is absolutely necessary, but a workflow and integration audit could be much more important.

I agree, doing a comprehensive audit of your current system is not as sexy as implementing the latest and greatest buzzword, but it can result in significant savings and efficiency and more effective patient care. Using “lessons learned” will provide a better ROI than applying “deep learning.” Remember this when next year’s buzzword is being introduced as well.

HIMSS Analytics introduces Imaging Maturity Model

Similar to the well-known EMR Adoption model, HIMSS analytics in Europe, in partnership with the European Society of Radiology (ESR), introduced at the 2016 ECR congress a digital imaging adoption model (DIAM) which promises to be a great model for institutions to benchmark their progress with regard to implementing a healthcare imaging IT strategy.  The model is quite comprehensive, it is defined using a scoring system of over 100 indicators from 10 focus areas, i.e. Software Infrastructure, Health Information Exchange, Workflow and Process Security , Quality and Safety Management, Patient Engagement, Clinical Documentation, Clinical Decision Support, Pervasiveness of Use, Advanced Analytics and Personalized Medicine. The model is currently being tested and piloted in several institutions in Europe in its initial phase. 

The different stages include the following:

Stage 0: Baseline, no electronic image management is present. Note that this does not mean that there are no digital modalities, but, rather, the information is not archived and actively managed.

Stage 1: Orders are electronically exchanged and available at the modalities. Images are exchanged and managed and reports are available and distributed electronically as well. All this is only on a departmental level, e.g. within radiology or a clinic.

Stage 2: The images are shared and available at the enterprise level for other physicians and specialties, for example through web access, or as a plugin on an EMR.

Stage 3: Workflow and process security is provided through status and change management so that the right images are available for the right patient at the right time. Quality measures are implemented such as peer reviews and critical result reporting and security and privacy controls such as audit trails are implemented.

Stage 4: Fully integrated image management at the enterprise level is provided through a Health Information Exchange, which can be private, i.e. managed by the enterprise. An enterprise can span multiple hospitals and clinics. Structured documentation is provided for measurements and other observations.

Stage 5, 6, and 7 include the following:

·         Advanced HIE functionality and Patient Engagement: A regional, typically a public HIE facilitates image exchange among practitioners, and patients are engaged such as through a portal.

·         Clinical Decision Support and Value based imaging: this provides feedback to a physician at time of ordering about the appropriateness of a procedure based on the patient preconditions, history and using practice guidelines.

·         Advanced analytics and personalized medicine: this allows for patient personalized or precision medicine based information to be used.

This model is a great tool for institutions as well as the user community as it will rank the level of imaging adoption. The next phase will potentially include non-European countries and include other non-radiology services. Hopefully it will be rolled out and take hold in the US and other countries soon as it can be a major differentiator between the various institutions.