Radiology AI & Diagnostic Innovation

Leonardo Kayat Bittencourt, MD, discusses how the UH RadiCLE is collaborating with innovative teams to use AI application to investigate clinical practices in hope of bringing new solutions to improve patient care and outcomes.

Radiology AI & Diagnostic Innovation
Featured Speaker:
Leonardo Kayat Bittencourt, MD

Leonardo Kayat Bittencourt, MD is Vice Chair of Innovation, Department of Radiology.

Transcription:
Radiology AI & Diagnostic Innovation

 Dan Simon, MD (Host): Hello, everyone. My name is Dr. Daniel Simon. I am your host of the Science at UH Podcast, sponsored by University Hospitals Research and Education Institute. This podcast series features University Hospitals' cutting edge research and innovations. Thank you for listening to another episode.


Today, I am happy to be joined by our guest, Dr. Leonardo Kayat Bittencourt. Dr. Bittencourt joined UH from Brazil as the Vice Chair of Innovation at the University Hospital's Department of Radiology. Dr. Bittencourt also leads the Radiology AI and Diagnostic Innovation Collaborative at UH, known as RADICAL.


Dr. Bittencourt is an expert in the MRI detection and staging of clinically significant prostate cancer. His clinical and academic career developed at Fluminense Federal University, where he has served as Vice Chair and Associate Professor of Radiology, and at DASA, where he served for 11 years as the Head of the Abdominal Division in Rio de Janeiro and Head of Prostate Diagnosis company wide. It's really a pleasure to welcome you here today, Leo.


Dr. Leonardo Kayat Bittencourt: Thank you very much, Dan. It's a pleasure and an honor to be here today.


Host: So Leo, early this year, the Department of Radiology at University Hospitals has partnered with the Health Systems Innovation and Commercialization Engine, UH Ventures, to launch a new effort that leverages UH radiologists research expertise in artificial intelligence applications. The Radiology AI and Diagnostic Innovation Collaborative, or RADICAL as you call it for short, led by you, aims to advance the science of teaching, research discoveries and clinical adoption of radiology AI while simultaneously serving as a revenue stream for the department and UH through its collaboration with outside entities. Tell us about this unique initiative. And in your view, what positions UH to be a lead in the field of artificial intelligence?


Dr. Leonardo Kayat Bittencourt: Thank you for the question, Dan. So, as you know, I arrived here in the institution by 2020 in the middle of COVID. The world was in lockdown and all we could do was have Zoom calls with people and try to connect and build a network by that. And as soon as I changed my job description from a Brazilian previous position to Vice Chair of Innovation, the invitations on LinkedIn, they started to pop up and asking me for that famous, like, 15-minute call. And most of those are trying to push their products to us for a trial. They wanted to sell things, but I immediately identified that many companies that were developing AI from outside of the US or that even they were within the US and they had emerging AI applications, they were also looking for partners to help them improve, validate or get their products through regulatory approval.


And the other thing that I immediately identified by joining the institution here is that we have a unique combination of a wealth of data, advanced applications, a diverse population of Northeast Ohio that we serve that is probably unparalleled in the country. World-class radiologists and specialists that are able to get all of those pieces together and bring any AI application to the next level, right? So, that drove interest from the industry and academic institutions and ultimately, by being connected with them, we're bringing benefit to our patients by introducing those new solutions to the improvement of their care.


Host: That's really terrific. You know, I think one of the things that'll really help our audience is to get right into one of your real success stories. So, one of the first projects initiated through RADICAL was with the French company AZmed. And AZmed had developed a tool called AI Rayvolve, which aimed to boost the speed and accuracy of fracture detection and solve the problems of missed fractures. UH is the only healthcare system in the US to collaborate with AZmed in testing this new technology. And the study done at UH played a pivotal role, in fact, in helping AZmed receive FDA clearance for Rayvolve to detect fractures. Can you tell us about this exciting project? I think a lot of us, you know, we're confused about the lay press. Is AI good, is AI bad? And this is just a great example of an amazing technology to help out emergency room physicians. So, tell us about it.


Dr. Leonardo Kayat Bittencourt: I fully agree with you. So, as you said, AZmed was our very first success story while building the program. And we were fortunate and grateful to have been chosen by them to be their first landing spot in the US for the validation of their product. And that required a huge effort also on the institution on our end, because there was only so much that I could do as an abdominal radiologist to help drive this forward. So, we had to build relationships and bring in other people; for example, Dr. Navid Faraji, from our musculoskeletal division who served as the leader, the clinical leader of this project and the PI for the whole clinical part of the study, and he's one of our AI clinical champions in the program now; also, the whole informatics in radiology that did all the data work around that and the research team, of course.


And that led to a study, a very complex study that was conducted here at UH that involved 27 UH providers as readers. So, imagine that enrolling 27 people to serve as readers, including MSK, so musculoskeletal specialist radiologists, fellowship-trained; also, general radiologists and non-radiologists, ED docs. So, combinations of those three groups, they read a number of different x-rays with and without fractures from different body parts. And after a washout period of a number of weeks, those people, they were offered again in a random order, the same pictures, but now with the aid of that tool that automatically detects fractures whenever they are present. So, there was a combination of positive and negative cases. And we wanted to measure then what was the percentage of improvement that we could find in the detection of fractures by the aid of that AI tool. And the results were really interesting and they speak very much to your introduction in your question that the ED doctors who are not radiologists, they had their initial sensitivity to fractures, which is around 80%, increased to 94%. The non-musculoskeletal radiologists, they had their sensitivities increased from a baseline 88%, so, like me, for example, as a non-musculoskeletal radiologist, to 96%. And the musculoskeletal radiologists who are already very well specialized, they did not have that much of an improvement because they are at the top of their game. So, it was a marginal increase between 92-97%. But the message is everybody, including the ED docs, who are not even radiologists, they were brought almost to the level of the highly specialized musculoskeletal radiologists. And that came also along with the 20% reduction in the time to read the scans, right? So, it's important to mention that it was time to read from when the scan was opened to when the read was finalized. So, that's a very minor, but significant improvement in timeliness too.


So, the next stage now is that after they got their FDA approval, which we're very honored and proud that we were a part of that. We are now in deploying in a pilot this application to clinical practice. And now, we are measuring the direct benefit to our patients. So, we are now collecting the data to identify if automated detection of fractures in patients presenting to the ED reflects to faster turnaround times overall for their x-rays to be read and a reduction of length of stay at the emergency department.


Host: You know, that's really terrific and I think it's a great practical solution to enhancing accuracy and speed of diagnosis in a place where both time and accuracy is really critical for patient care. Tell us a little bit about the same efforts in pneumothorax detection following line placement in intensive care units and throughout the hospital.


Dr. Leonardo Kayat Bittencourt: Yeah. And that's also very, very impressive. I don't have the accurate numbers at the moment to share, but we are experiencing significant shortening of the time to read the scans of patients who even who have unsuspected pneumothorax. So, patients that have non-STAT exams that are ordered on x-rays and unsuspected pneumothorax that they could have been read as routine scans, they are now moved to the top of the list and they are notified as soon as possible. And also, the patients who have suspected pneumothorax are read fast. So, Dr. Amit Gupta who is our Division Head for the Cardiothoracic Division of Radiology. He has extensive experience in chest AI already with multiple applications already in clinical practice, including not only pneumothorax detection, but also endotracheal tube positioning. So, the department is moving in every direction to fully embrace the power of the augmented abilities that AI is bringing to radiology.


Host: So, I guess the $64,000 question, as we say is, will we still need radiologists or will AI replace you? I suspect that you're going to tell me. That it's going to help you be better radiologists.


Dr. Leonardo Kayat Bittencourt: I fully agree with you. And I always cite this citation that I heard from Dr. Hricak, Former Chair of the Memorial Sloan Kettering. She said that if you're not at the table, it means that you are on the menu. So, we need to embrace those kinds of discussions. We do not need to shy away from them because they seem like a threat. If we are not an active part of that, then it means that we are losing our relevance in that set. And I believe that by doing so, by stepping up to the AI game, UH is gaining national and international relevance as an active player in the decisions related to adoption of AI. So, I don't think that radiologists or any other specialties will be replaced by AI. But again, as other famous people like to say, radiologists or practitioners that use and are augmented by AI will replace those who refrain from doing so.


Host: Terrific. So, UH Radiology joined Massachusetts General Hospital and other prestigious academic medical centers as a charter member of a consortium convened by the American College of Radiology to work on the validation of AI algorithms. The first project validated a stroke detection algorithm in both CT and MRI, along with few others including improving detection of prostate cancer in MRIs and opportunistic screening of cardiovascular risk from routine x-ray-based imaging exams. Please share with us about this consortium and research projects through this effort that may lead to new ways to benefit patients.


Dr. Leonardo Kayat Bittencourt: This is very exciting. So, the general concept around anything in AI is that more data, more diverse and more robust data leads to better AI and more generalizable AI. So, it makes no sense if we only collect data from a very narrow extract from the population and think that we can use throughout the whole population that we serve.


So, that's the reason behind those consortiums and those multi-center initiatives. And we were also fortunate and lucky to be included in this consortium from the ACR, including other institutions such as MGB, that you said. And the first use case was to collect cases to validate a stroke detection and characterization solution that was being developed by a commercial company at the time. And the importance is that by joining this consortium, we are not only getting access, of course, in the academic life to publications, to more collaborations and potential grants and so on, but we also have access in the future to the validated applications once they are available. And in case we decide when we want to use the data from the whole consortium for other use cases, we can always apply to use this data and it's much easier to leverage on that than to collect the data from scratch.


And in the end, from the marketing standpoint is also important because you're joining a more select and exclusive club. And your brand and your name and your institution, it gets more and more associated with other leaders in the field and well-recognized by that. So, that was one of the first consortiums to which we had access, but now we are opening the doors and joining other similar initiatives also as a pipeline and as a sales channel to more similar opportunities.


So, I can bring as an example, a company called Bunker Hill, which is in itself a consortium, that their mission is to bring together different agents along the AI development pathway. So for example, if you have an academic researcher that wants to develop an algorithm, but they need data, so they don't have access to data, so they match the researcher with people who have the data. And then, you need someone to annotate and ground truth and prepare the data, so to the curation of the data, they match with those people too and they bring all those people together and they share the profits of those applications, whenever they are commercially available, commensurate to the participation of each one. And we have been already successful in at least three projects with them. And I would bring as an example, one that was led by Dr. Amit Gupta, again, always a big champion in chest AI imaging. And it was a tool, if I'm not mistaken, that aimed to calculate the calcium score of coronaries, so your area of expertise, over routine CTs of the chest. And they got the FDA approval of this project. And UH is going to get now a license to use this commercial tool for free and to the benefit of our patients again. So, there are different things that we can take away from the participation in those consortiums. And it's always reflecting to our mission to care for our patients and improve their outcomes.


Host: Well, that's really terrific. So, that's a national consortium. But as we know, medicine is global. It's one of the reasons why we recruited you from Brazil. So early this year, UH Radiology has entered into a new research and development partnership with DASA, the second largest private healthcare system in Brazil and the largest diagnostic medicine company in Latin America. DASA is also Brazil's leading institution in radiology search and a world leader in digital innovation and artificial intelligence. This is a very exciting partnership which may lead to great opportunities for joint research projects and new technologies and applications. Can you tell us a little bit more about this collaboration? And obviously, You work there, no one knows more about it than you do.


Dr. Leonardo Kayat Bittencourt: Sure. And I'm, very proud to talk about DASA and that part of my story and the amazing friends, mentors, and almost the family that I have there still today. So in terms of DASA's big numbers, we're talking about a company that encompasses more than 10 states of Brazil. So, Brazil is also a continental country. They have more than 900 care units throughout the country, including hospitals, outpatient centers, primary cares, labs, and so on. They employ more than 3,000 physicians only and more than 40,000 employees, covering 14 million patients and performing, processing 250 million exams a year. So, we can imagine now the amount of data that they generate and also, similar to us, well, at a different scale probably, but also the diversity and the complexity of the operations under which they perform.


And in terms of the academic achievements, even though they are a private company, but in Brazil, there is a situation that we can have dual appointments, and the private supports the public sector in Brazil in producing research. So, people that work at DASA, they are responsible, historically, for over 50% of radiology research in Brazil, right? And I think that we can say the same for clinical pathology and pathology too. They have a huge footprint in clinical AI, with numbers of imaging and non-imaging based algorithms deployed to clinical practice. And they are one of the world benchmarks in terms of operational efficiency. They always win the awards of the most productive MRI scanners in the world or the Leanness production line. So, there are a lot of things that we can learn from one another and that we can exchange. So as a result of this master collaboration, we expect to leverage on joint publications to collaborate on data for different kinds of projects, including AI, but also in research to offer our program, our RADICAL Program in our radiology department and our UH as a US validation site for whatever AI tools they think of developing there, and also to offer, and that goes to the whole UH community. Anybody interested in establishing a program for visiting trainees and professors back and forth, I'll be happy to help people broker that. And we are happy to announce also that we are going to be hosting a visit from their leadership at the end of this year to continue this partnership. So, this is something that makes us very happy and proud and we hope that can bring results to everybody here in the UH community.


Host: Well, Leo, it's been so great talking to you today. It seems almost like yesterday when I was just interviewing you in my office to try to convince you to move from sunny Rio to shady Cleveland, cloudy and shady, not the same sun. But I'm so glad that you're here. Thank you for taking the time to speak with us today, Dr. Bittencourt. For our listeners interested in learning more about research at University Hospitals, please visit uhhospitals.org. Thank you.