Bats Out of Hell
SIIM23 Technology Hype Curve
I had to think about this 1977 record by Meat Loaf, which sold over 43 million copies worldwide (I was one of those 43 million fans who played this record over and over again), while watching the bats coming out from under the Congress Street bridge in Austin, Texas, while attending SIIM23. Actually, the bats were a no-show as there were over 100 spectators, and I waited for two hours to see them, to no avail. Apparently, if they don’t like the weather conditions, they prefer to keep on sleeping.
Unlike the bats, who were a no-show, more than 700 attendees were at the SIIM23 conference while having more than 80 exhibitors. The intimacy and closeness make this event so convenient, unlike other events such as HIMSS, RSNA, or even ECR, as everyone can easily approach and reach the halls and conference floor.
Here is my SIIM23 technology hype curve based on the “infamous” Gartner hype cycle curve based on my observations:
- Technology trigger.
- ChatGPT: This technology is still in its infancy and hype phase. Presenters quoted publications showing that AI-created summaries of diagnostic reports are still not as good and complete as if a physician generated them. Even AI-generated synopsis for patients, which would “translate” a diagnostic report in sixth-grade level readability, would still miss some information. There are good opportunities in the short term for translating into another language, and if checked carefully by a physician, it could be helpful. However, the fact that it still requires significant human oversight defeats the purpose and promise of saving time and providing higher productivity. It was also interesting that patients perceived AI-generated text and responses as being more emphatic than those coming from a physician. Overall, a promising technology, but still in its infancy.
- Virtual Reality: This technology is not at the peak of the hype curve. I “visited” a virtual reading room that can be used to invite others. The image display in the goggles is calibrated to provide diagnostic-quality images. I still had to use physical hand controls, which hopefully will be replaced by gestures such as demonstrated by the recently introduced AppleVison headset. It can be an excellent tool for training and education, but large-scale commercial adaptation and deployment are still quite a way off.
- Peak of inflated expectations.
- FHIR: Although SIIM is imaging oriented and most of the FHIR deployment is on the periphery, I typically do an informal poll on who deployed FHIR in their interfaces. Unfortunately, I have found very few providers or vendors with any FHIR interfaces deployed. One large teleradiology provider, who supports 6000+ interfaces over thousands of their sites, told me they tested their interfaces with the respective “FHIR sandboxes” provided by the EMR vendors but have not gotten any takers on an FHIR interface. One informatics infrastructure vendor told me they deployed one resource (Patient) to get MPI data, but only after a painful integration. Remember that FHIR has 100+ resources defined, so one represents one percent. There appears to be little incentive to move away from HL7 V2, and there is a fear of being on the “bleeding edge” despite all the hoopla around this new standard.
- Trough of Disillusionment.
- AI: The good news is that AI is over its hype peak, but the bad news is that it is still not quite “enlightening.” Many algorithms do not perform as promised, primarily due to a lack of non-biased training data. There is no clear path to integrate these algorithms into the workflow because:
- Some PACS vendors use a somewhat restricted marketplace,
- Some providers select to use a neutral AI platform they can manage themselves,
- Some do a direct API integration, and
- Some rely on the Cloud.
Current U.S. regulatory approvals are not keeping up with the fact that a single AI application can detect different conditions, unlike in Europe and other continents. Incorrect AI detections seem to bias physicians’ decisions, and last but not least, patients do not feel comfortable being diagnosed by a “computer.” AI has a lot of promise and potential, but we are not quite there yet.
- AI: The good news is that AI is over its hype peak, but the bad news is that it is still not quite “enlightening.” Many algorithms do not perform as promised, primarily due to a lack of non-biased training data. There is no clear path to integrate these algorithms into the workflow because:
- Slope of Enlightenment.
- Enterprise Imaging: Images are now available ubiquitously inside the healthcare enterprise. However, there are some caveats. First, image availability applies primarily to radiology imaging. Many institutions haven’t incorporated their other specialties, such as dermatology, ophthalmology, surgery, and definitely not pathology. The biggest stumbling block is the workflow, i.e., ensuring that the images are properly tagged with the correct patient identification so the PACS/VNA can appropriately manage them. Tagging is crucial with the massive influx of point-of-care imaging devices, as physicians can take pictures with their phones and perform Ultrasounds with a simple probe attachment to their phone or tablet. In addition, many CDs are still created; therefore, the open infrastructure to exchange images among institutions (at least in the US) is still not quite there. So, there is still much work to do, but we are getting there.
- DICOMWeb: If some excellent presentations on this topic were missed, they are still available through SIIMTraining. This new protocol has many advantages using web services instead of the conventional “DIMSE” protocol. Especially now, as most EMRs can view images directly from their portal, viewing from a cloud resource is becoming a more popular, web-based image access methodology starting to proliferate. There are still a few kinks to figure out, but DICOMWeb deployment is beginning to become enlightening.
- Plateau of productivity:
- EMR/IHR: EMRs are ubiquitous. EMRs are less intuitive and efficient than many physicians would like, but it has become the primary tool for almost every physician. An interesting discussion was stated by Florent Saint-Clair of Dicom Systems, showing that the Imaging Health Record is a companion to the EMR in trying to elevate its importance at the same level since many CIOs are spending more time on EMRs than could be justified by its importance. Remember that an image is often more valuable than 1000 words. Hopefully, this will change the perception of the healthcare IT community and physician communities’ perceptions as they become less reliant on the reports and get more access to the true source data, i.e., the images.