Dr. Ramsey M. Wehbe Wins ASNC/Pfizer Young Investigator in Cardiac Amyloidosis Research Fellowship Award
ASNC congratulates Ramsey M. Wehbe, MD, MSAI, recipient of the 2021 ASNC/Pfizer Young Investigator in Cardiac Amyloidosis Research Fellowship Award. Dr. Wehbe will receive $50,000 to support his research proposal, “Validation and Implementation of an Artificial Intelligence Augmented System for the Diagnosis of ATTR Cardiac Amyloidosis on Cardiac Scintigraphy.”
The ASNC/Pfizer Young Investigator in Cardiac Amyloidosis Research Fellowship Award was established through a grant from Pfizer, Inc. The award is part of ASNC's broad effort to expand opportunities for clinical research in cardiac amyloidosis and improve the outcomes of patients with ATTR cardiac amyloidosis. Each year's awardee is an early-career researcher whose proposal has the potential to advance both the science of cardiac imaging and the nuclear cardiology specialty.
“We received excellent proposals in competition for this award, which underwent careful peer review by experts in the field,” says Albert Sinusas, MD, FASNC, chair of the ASNC/Pfizer Young Investigator in Cardiac Amyloidosis Research Fellowship Award Selection Committee. “Dr. Wehbe's proposal was considered outstanding and highly innovative involving the design and application of a deep convolutional neural network architecture in the evaluation of PYP cardiac scintigraphy. The committee also felt that Dr. Wehbe had a unique technical and medical expertise, having completed a research fellowship in artificial intelligence and in clinical cardiology fellowship with ongoing training in advanced heart failure. His work will likely have a significant impact in the diagnostic evaluation of cardiac amyloidosis.”
Following the presentation of Dr. Wehbe's award at ASNC2021, we had the opportunity to ask him about his AI research and learn how his new project could improve cardiac amyloidosis diagnoses. Here are excerpts from our conversation:
ASNC: Congratulations on winning this year's ASNC/Pfizer Young Investigator in Cardiac Amyloidosis Research Fellowship Award. We understand your project will evaluate the performance of a new deep learning system for detecting ATTR cardiac amyloidosis on cardiac scintigraphy. Would you tell us a bit about how the deep learning system works and how it might help cardiologists in the future?
RMW: Thank you very much, and thanks for the opportunity to highlight this project, which I am incredibly excited about given its potential to directly benefit patients being evaluated for ATTR cardiac amyloidosis.
ATTR cardiac amyloidosis remains an underdiagnosed cause of heart failure and carries with it significant morbidity and mortality. Early diagnosis is key given we now have an effective therapy that improves survival and quality of life for these patients, the TTR stabilizer tafamidis.
Thankfully, we have also had remarkable advances in diagnostic tools for ATTR-CA, particularly cardiac scintigraphy with technetium pyrophosphate (“PYP imaging”). Cardiac scintigraphy has become a key component of the noninvasive diagnosis of ATTR-CA with reported performance in the literature similar to the invasive gold-standard of endomyocardial biopsy.
Unfortunately, we and others have observed that the “real world” performance of cardiac scintigraphy for detecting ATTR cardiac amyloidosis outside of expert centers may be much lower, as there are many potential pitfalls in the interpretation of cardiac scintigraphy for the inexperienced clinician. This limits the widespread clinical utility of this important diagnostic modality and may lead to unnecessary invasive follow up testing or missed diagnoses.
One potential solution lies in leveraging advances in artificial intelligence. Deep learning is a form of artificial intelligence based on artificial neural networks inspired by the human nervous system and is a revolutionary technology underpinning innovations like self-driving cars and smart voice assistants. A remarkable characteristic of deep learning is the ability to model complex, unstructured data like images without explicit instructions or programming, just by exposing the model to labeled examples. This is the same way a child, for instance, learns to distinguish between a cat and a dog — by being shown many examples and learning from mistakes. Similarly, we can train a model to learn to distinguish between positive and negative cardiac scintigraphy studies for ATTR-CA by simply exposing it to hundreds of labeled examples.
We have built a pilot deep learning system for the detection of ATTR-CA on cardiac scintigraphy with performance on par with experts at our institution, and we are currently refining this system further. Our hope is that this system will help “level the playing field” and allow for earlier and more accurate diagnosis of ATTR cardiac amyloidosis even in resource-constrained settings where an expert extensively trained in the interpretation of cardiac-amyloid-specific radionuclide imaging may not be available.
ASNC: Your proposal emphasizes that validation across different practice types is necessary to move the deep learning system forward How will your ASNC/Pfizer award further the validation effort?
RMW: Absolutely. While we have seen a lot of promising research using deep learning to solve problems across medical specialties, there is still a huge implementation gap, and very few of these systems actually make it to the bedside to impact clinical care.
One of the largest barriers to clinical implementation and widespread adoption of these technologies is poor performance outside of the patient cohorts on which they were developed. This lack of generalizability is not necessarily unique to deep learning but, due to their complexity, deep learning systems in particular are prone to overfitting to the data on which they were trained. This problem is akin to memorizing the answers to a test rather than learning the underlying concepts.
Even more concerning, a model that overfits to bias in data can lead to systematic bias or prejudice in its predictions. It is therefore imperative that any deep learning system intended for clinical use is repeatedly validated in diverse cohorts outside of the institution or dataset on which it was trained.
In addition to prospective validation across the Northwestern Medicine system, the ASNC/Pfizer award will allow us to partner with other health systems to rigorously validate our deep learning system. We also plan to study creative ways to improve the generalizability of our system to other institutions using novel methods so that we can ensure the system benefits patients regardless of demographics or where they receive their care.
ASNC: Your proposal also includes creating a framework for integrating the deep learning system into the nuclear cardiology workflow. That seems like an exciting leap forward. Can you talk about how nuclear cardiologists might be able to use this new AI tool in day-to-day practice?
RMW: Yes! Aside from generalizability, a major barrier to putting a deep learning system into practice is the ability to assimilate the system into the clinician's workflow. An effective system improves rather than hinders efficiency.
Additionally, for clinical acceptance, a deep learning system should have some degree of explainability or interpretability built in to facilitate trust in algorithmic predictions as well as transparency when there are obvious errors made by the system. This human–machine interface is key since there is evidence that, when implemented effectively, predictions from a human working in concert with a machine are superior to those from a human or machine alone, a concept known as augmented intelligence.
Our goal is to work with human–computer interface experts at our institution as well as partner with vendors of nuclear imaging software to integrate this system directly into the nuclear imager's workflow so that we can bring this technology to the bedside.
ASNC: Thank you giving us a glimpse into your work and its exciting potential. The cardiovascular imaging community will look forward to your results presentation at ASNC2022. Is there anything else you would like to share?
RMW: I want to thank the American Society of Nuclear Cardiology and Pfizer, Inc., again for the generous support of this research. I also want to thank all of my incredible research and clinical mentors at the Northwestern University Feinberg School of Medicine who have made this work possible, including my co-mentors for this project, Dr. Sanjiv Shah and Dr. Aggelos Katsaggelos. Finally, I want to acknowledge all of the patients who are living with ATTR cardiac amyloidosis. My hope is that this research leads to earlier and more accurate diagnosis of this condition so that patients can benefit from life-prolonging therapies before the disease has progressed to an advanced stage.
Tweet your congratulations to Dr. Wehbe at @ramseywehbemd.
For more information about the ASNC/Pfizer Young Investigator in Cardiac Amyloidosis Research Fellowship Award, visit ASNC.org Awards.