Now Is the Time
Over the past few years, we’ve seen incredible advances, personal sacrifices, and humbling moments in the healthcare and biotech industries. Hero’s working tirelessly to save lives during a global pandemic, adaptation to unnatural societal guidelines, and record-breaking scientific discovery for vaccines that have offered global protection against an unseen virus. Yet with all our advances, there are aspects of healthcare that continue to be stalled in a spot that has plagued our industry for years. In terms of data – and data sharing – Healthcare has historically maintained a siloed culture, and we still have a way to go to improve information exchange and care continuity. After all, a patient’s journey is not complete, until we merge their clinical imaging records with their electronic health information.
When observing the operational aspects of healthcare workflow processes, there continues to be a lot of opportunity to streamline operations and workflows, which will improve productivity and patient care. Disjointed processes can affect clinical team collaboration potentially causing a breakdown in patient care. I frequently see articles that discuss interoperability challenges in healthcare, yet the same challenges exist today as we experienced in the past. When Meaningful Use initiatives were first introduced, bringing about widespread adoption of EHRs, focus was given to meeting government mandates, not necessarily user ease. This caused extra work to already overworked healthcare professionals from routine and mundane tasks in various workflows affecting employee morale and productivity. In recent years, however, EHR software releases are more focused on end-user workflows, and there continues to be vendors developing solutions to fill unmet gaps in workflow processes.
In the United States, healthcare spending represents around 10% of our GDP and we spend more on healthcare than any other country. We’re at the right time to redefine how healthcare is practiced, using technology that is advancing almost at a daily pace. In the next 10 years, I predict we will have all patient information interconnected and available at our fingertips when needed, turning a long-term vision into reality.
Redefine Radiology – Again?
Radiology has “redefined” itself several times throughout the decades. Looking back, in 1971 the first CT scanner was introduced, ushering in the digital age of imaging. Although film would be around for some time, (it took almost 20 years before the rapid transition to digital happened) the next redefinition of radiology was around 1990, as organizations began the move to Picture Archive Communication Systems (PACS) and replacement of film-based to digital acquisition devices. For those of us who participated in the transformation from film to digital, it changed how radiology was practiced and it’s never been the same! Digital technology ushered in new modalities that advanced our clinical understanding of various anatomical systems, and sophisticated artificial intelligence (AI) and machine learning tools provided another advance in disease detection and workflow automation. The potential of AI in advancing medical imaging is truly quite fascinating.
After the introduction of PACS, the next 10-15 years saw tremendous growth as global transition to digital technology became the norm. Along the way, new integrations to PACS improved radiology workflows and significant speed in image delivery became paramount, but speed of image delivery also created new workflow challenges that would need to be solved, as well as opened new avenues for improving workflows to speed result turn-around times.
Yet with all that change, the acquisition side of imaging still remains basically unchanged, and while technology has helped improve image acquisition, quality, and transmission, the overall workflow of radiology provides the greatest opportunity for improvement.
Artificial Intelligence (AI) – Healthcare’s New Disruptor
If you think about it, the journey of radiology has led us to this moment, and probably the COVID pandemic helped push AI to the forefront even faster.
AI’s history is quite interesting; starting in the early part of the 20th century, a “computerized robot” concept started the AI wave. Around 1950, Alan Turning writes Computing Machinery and Intelligence followed by an AI project at Dartmouth, led by John McCarthy and Marvin Minsky. From the late 1950s to the mid-1970s, computer costs became cheaper as researchers learned that the algorithms they had to develop needed more processing power, stagnating development for a time. During the 1980s, AI gets rejuvenated when computers become more robust and accessible, and funds start to be allocated to AI development. In 1997, Gary Kasparov – world famous chess player – defeated IBM’s Deep Blue, and around the same time, Speech Recognition was introduced. In the 1990s-2000s, Moore’s Law applies to mainstream computing, meaning memory and speed doubled about every year.
AI in healthcare, started around 1959 as Robert Ledley and Lee Lusted published a paper about the importance of reason processes in medical diagnostics. In 1965, AI was gaining attention at Stanford University, and the National Institute of Health created an AI laboratory in conjunction with Sanford. In the 1990s, IBM released Deep Blue and Watson. Then, around 2020, Google’s DeepMind uses AI to solve the protein folding problem (a perplexing challenge that existed for over 15 years).
AI’s history has had a lot of starts and stops but today, we’re in the mainstream of computing which allows us to create more sophisticated algorithms every day. A few years back, it was noted that AI will replace radiologists in imaging diagnostics and that statement sent shockwaves throughout the medical imaging community. Fortunately, we are now confident that AI is a tool radiologists will use to aid them in diagnostic accuracy, not replace them altogether. Why? Because computers cannot evaluate patients, whereas physicians can incorporate patient conversations and observations, physical interactions, medical history, and clinical content to arrive at a diagnosis.
AI will continue to evolve and solve problems that have plagued our industry for decades (yet most are ignored because we’ve found “workarounds”). Let’s consider a few of the opportunities AI has in radiology that provides significant benefits:
Wander through the aisles at major imaging conferences and you will find many AI vendors touting their unique algorithms to detect various abnormalities. These algorithms are gaining trust and becoming valuable tools given various diagnostic situations, but most are “point” solutions. Each algorithm is created to detect one condition based on their FDA approvals, so most organizations may have multiple “point” algorithms based on their patient population and workflows. Each one must become part of the workflow, and all require time to render a diagnosis. They are useful in medical diagnosis because the AI may detect subtle nuances in pixel data not easily detectable to the human eye, and relay that to a radiologist for further review.
Prioritization of Urgent Cases
AI can significantly streamline workflows in medical imaging. With the increase in volume and the shortage of radiologists, having tools available to help “triage” images to the appropriate radiologist is a huge time saver. AI can detect abnormalities vs. normal studies and route the normal exams to residents (or teleradiology) to complete a report and route the abnormal images to radiologists based on specialty. This can significantly improve workflow and ensures the right exams make it to the right specialist at the right time. This also has an upside of improving patient care and outcomes.
Eliminating redundant, mundane, and menial tasks is another area AI adds tremendous value in medical imaging. Observing physician and technologist workflow while using technology, you will probably identify tasks that could be accomplished by a computer, not a human. One example is ensuring you have consistent data standards in your DICOM files to ensure hanging protocols display properly. This doesn’t need to be handled by the modality, tech, or PACS; let AI evaluate each image and data set and apply a standardized nomenclature to all your images. This prevents your radiologists from having to rehang or manipulate images after they load a study, saving them valuable time and mouse clicks, while greatly improving their productivity.
Imagine looking back through your image archive and analyzing your medical images to understand your mix of cases and outcomes, diagnostic accuracy (were studies over read or under read when initially reviewed), what potentially was not reported on that could have been used for additional billing purposes, and then cataloging that information from your images as well as the report information into a real-world database for application with future studies. Furthermore, providing insight on past unbilled opportunities provides an opportunity to improve billing accuracy (and future revenue).
This should probably be the number one benefit for AI in healthcare next to diagnosis. For any AI algorithm to work properly, your data needs standardized, otherwise your models could fail at a much higher rate than expected – and building a real-world databased on non-standardized data will not work. Improved hanging protocols is another obvious benefit of data standardization which improves radiologist productivity. Data standardization makes interoperability with 3rd party applications and data exchange among facilities successful and much easier. Finally, standards allow you to capture key information, store it in a database, and apply insight to that information on future diagnostic cases, and that can help to improve patient outcomes.
Healthcare is continually faced with new challenges and incorporating AI into your workflow is going to become mainstream for almost every radiology department. One thing that will become critical to assess as you embark on your AI journey, is figuring out how to manage all the point solutions you may acquire in your organization. Considering a comprehensive data platform that can manage data standardization, workflow orchestration, system integrations, and overall management of all your point AI solutions makes the most logical sense, as this type of AI platform can significantly reduce the number of integrations and IT overhead required for multiple point solutions.
AI strategy is as important as the AI technology itself. As you embark on your AI journey in your healthcare system, there are some things that are important to plan for. Next month, I share some tips for developing an AI strategy and adoption.
Finally, if you would like to learn more about how Enlitic is changing healthcare by solving high value problems and can provide you with a platform to manage your AI investments, including data standardization, workflows, and our own advanced AI clinical tools, check out our website or reach out to me directly to start a conversation!