What Does AI Mean to You? – Part 2

Abstract

Now that we have established that AI does not refer to valve regurgitation (see my May editorial), let’s take a few minutes to investigate the potential impact of Artificial Intelligence on our daily (echo) lives. There is a growing level of enthusiasm for the term Augmented Intelligence, recognizing that AI in medicine will never be fully isolated (e.g., totally Artificial) but will instead be used to augment physician input (Augmented Intelligence definition: “Design pattern for a human-centered partnership model of people and artificial intelligence (AI) working together to enhance cognitive performance”1Gartner Glossary.https://www.gartner.com/en/information-technology/glossary/augmented-intelligenceDate: 2022Date accessed: May 15, 2022Google Scholar). I once heard on National Public Radio that we check our phones an average of 52 times each day. I wasn’t sure if I was supposed to be shocked by that frequency or not, but in fact, I immediately thought: “Is that all?” As a means of primary communication as well as a direct connection to a global information network, I imagine I would fall to the far right-side of that bell curve (assuming a normal distribution). The point I am making here is that we all carry sources of help at our fingertips; how we use this help is highly variable. I think AI will provide similar support that will be exploited to a variable degree. AI is a very generic term but serves the purpose of offering us an over-arching name that simplifies an otherwise overwhelming and potentially offsetting group of terms and abbreviations (e.g., NLP, ML, bioinformatics, blockchains, crowdsourcing, EMRs, LHS, clustering, & more). AI is a disruptive technology, with disruptive terminology, that involves the use of computerized algorithms to dissect complicated data. By way of background, “artifice” is defined as: “Cunning device or expedient, especially as used to trick or deceive others.”2Lexico.https://www.lexico.com/en/definition/artificeDate: 2022Date accessed: May 15, 2022Google Scholar Based upon this root definition of artificial, I support the transition to AI meaning Augmented (or collaborative) Intelligence as a more accurate term, with our stated goal of incorporating advanced analytics and computing power together with our personal responsibility of interpreting its outputs to become more intelligent in the care of our patients. If you have not already done so, you need to embrace AI as something that is here now, has completely altered the cardiovascular landscape forever, and will continue to impact how you practice medicine. AI facilitates detection and quantification of images. With limited effort, I can recite many instances where computer-based tools are influencing your day-to-day activities (e.g., digital tracking from wearable technology; asynchronous communication from electronic consultations; virtual visits; eICUs; automated border detection for SPECT reporting; automated segmentation in 3DE, CCT, and CMR, and the very recently approved use of AI for LVEF and GLS). AI is designed to solve problems. I believe that echocardiography is currently approaching an historical set of fundamental problems. Mounting pressures from unlimited patient volumes coupled with an unprecedented reduction in our work-force; seemingly never-ending expanding imaging capabilities matched to less robust scientific scrutiny of their clinical value; an unparalleled list of alternative non-invasive diagnostic imaging options (including the continued changeover of cardiac ultrasound to non-cardiology users); and a general societal sentiment that contemporary technology supplants conventional medical products, seem poised to stockpile and fire the cork out of the echo bottle. We need a sustained effort to piece together AI capabilities with these expanding problems to provide long-standing solutions. In my lab, if AI could guarantee that the studies performed today were appropriate (integrating clinical, laboratory, and other imaging results), prioritized based upon the needs of the patient (rather than the needs of the hospital administrators), analyzed using sophisticated quantitative parameters (instead of an eye-ball review of risk-predicting chamber dilation or valve severity), and matched to clinical outcomes (through the incorporation of all EMR data), I would fully embrace AI as a major development with long-lasting impact. As of the writing of this editorial, I have fully embraced and adopted the concept of AI. However, to fully embrace the implementation of AI and operationalize this tool in our global imaging environment, each of us will need to actively participate in its development. For example, sonographers are skilled professionals, but we ask them to perform menial tasks on a daily basis. AI can and should address their daily practice to make them more efficient. How difficult would it be to have AI embedded within our ultrasound systems and responsible for tracing all Doppler spectra, averaging multiple beats, and reporting their clinical meaning based upon rules established by the American Society of Echocardiography’s guidelines and consensus statements? Seems like a low-bar request for bioengineers, and we should be making these types of requests now. Even more palpable, if AI could proof-read every study, place all normal reads toward the end of my reading stack, and alert me to all reporting discrepancies before I sign off, that would impact my day-to-day practice in a pragmatic way. Furthermore, and a recent area of intense investigation, AI can help find patients that would benefit from further care. I ask you: Do you know the number of patients with severe symptomatic aortic stenosis that did not get a cardiology referral after you reported their life-threatening pathologic valve finding? AI is now being used to assist in making those connections in a retrospective manner by connecting the echo report with the EMR data systems. Taking this a step forward, your message to the referring physician requesting referral to the valve clinic should occur in the background with no additional effort on the part of you or your staff. What if AI was able to assist you by offering greater precision in phenotyping? Imagine the ability to risk stratify a very diverse disease state such as HFpEF. Deep learning algorithms could enhance the assessment of imaging features such as echo-texture, volume, and shape to potentially augment the physician’s ability to diagnose the etiology of LV wall thickness and reduce the need for additional imaging or invasive cardiac biopsies in established low-probability cases. Can we harness the potential of AI to identify imaging patterns associated with hard clinical outcomes (e.g., severe arrhythmias, impending hemodynamic instability, event-specific mortality) rather than reporting non-specific echo parameters typical of echocardiography practices in 2022? That would certainly increase our excitement for reporting diastolic dysfunction. In this month’s issue of CASE, you will find rare, but nearly pathognomonic, images that could be included into AI algorithms for future system-level recognition for end-users not familiar with these phenotypic patterns (Arow et al, Cotella et al). You will also find a number of new Journal Categories emphasizing the continued growth of the Journal. These include Critical Care Echocardiography (stressing the important role of serial examinations in decision making), the Hemodynamic Corner (focusing on physiologic maneuvers, waveforms, and correlation with invasive pressure tracings), and a potpourri category I have simply titled Just Another Day in the Echo Lab (highlighting the varied clinical presentations that bring patients to our everyday attention). Examples of effusive-constrictive pericarditis (Haq et al), platypnea-orthodeoxia syndrome (Mima et al), non-bacterial endocarditis – where severe AR was treated with anticoagulation alone (Murata et al), and an eye-popping example of what can go wrong during ECMO for cardiogenic shock (Anastasius et al) should be good educational tools for you to use during teaching conferences and lab meetings. To round things out, Rodríguez-Pérez et al illustrate how the RV strain pattern may be used to guide response to therapy in patients with IPAH, and Binder et al demonstrate the value of TEE to determine if a percutaneous or surgical treatment approach should be employed in a patient with a large, complex ASD. Given what we anticipate AI will one day deliver, it is easy to appreciate where AI may have impacted many of these CASE examples. In general, the implementation of AI into daily clinical practice requires an adoption, an implementation, and an operational phase. In my opinion, most of us could benefit from taking this perspective back to our own think-tanks where we can define our most pressing problems and ask how AI can address those (e.g., matching the detection of disease with appropriate referral or management strategies). During the initial phase of COVID-19, it was AI that addressed an important solution to a defined problem and provided the opportunity for nurses and frontline practitioners without specialized training in cardiac ultrasound to obtain diagnostic quality POCUS images. This work was highlighted at the ASE 2020 Virtual Experience; you can read more details here. So, I ask you: Are you prepared for AI in your lab? Without this intelligent ally, how will you meet the increasing demand resulting from more patients in need of echocardiography as obesity, hypertension, and diabetes remain rampant and the aging population boom nears its peak? Our Society is taking the lead on matching the needs of our echo labs with the capabilities of AI. Through the AI Forum, think tank programs, the AI Standardization subcommittee, and the creation of the ImageGuideEchoTM Registry, the ASE is assisting in the creation of the algorithms that will soon be incorporated into all of our ultrasound platforms. It is this critical oversight, beyond industry, that is necessary to prevent a serious dilution of the qualities offered by echo that have matured over the past 60 years. Each of us serves an important role to offer consistent global scrutiny of this process to prevent echo from being dumbed down to an unusable tool, making it even harder than it already is for echo-lab-outsiders to recognize its phenomenal value. I’ve heard the following quip recited many times and I tend to agree with this sentiment: AI will not replace a doctor; but a doctor who uses AI, may replace a doctor who does not use AI.

Ähnliche Arbeiten