Meet the Northwestern leader developing AI for clinical operations – Becker’s Hospital Review | Healthcare News
As AI technology increasingly amalgamates into the medical arena, health system leaders are stepping up to help steer the path forward.
Women comprise only 30% of those working in AI today, and Adrienne Kline, MD, PhD, is one of them. She is head of artificial intelligence and engineering at Chicago-based Northwestern Medicine Bluhm Cardiovascular Institute and principal investigator of the Kline Engineering Lab. She has received $1.4 million in research grants to develop AI technologies aimed at improving both patient care and clinical operations.
Dr. Kline shared insights with Becker’s about the clinical tools she and her team are developing, and what health system leaders should have been doing with AI — yesterday.
Editor’s note: Responses have been lightly edited for clarity and length.
Question: How is your latest AI-based tool, Project Corazon, designed to support cardiac diagnostics more precisely? What operational benefits does it bring to cardiovascular care delivery?
Dr. Adrienne Kline: Project Corazon harnesses the power of cutting-edge AI technologies, specifically trained on large cardiac imaging datasets, to efficiently deliver comprehensive and accurate reports specific to the needs of clinical diagnosis. Previous attempts have used black-box generative approaches, but Corazon was built from the ground up to replicate the cognitive and reasoning processes used by expert cardiologists. By automatically identifying relevant features and components of cardiac anatomy, it can then directly measure ejection fraction, detect wall‐motion abnormalities, all of which feed into diagnostics.
Rather than replacing physicians, Corazon’s architecture augments their decision-making while reducing inter‐observer variability and even flagging early signs of disease that might escape routine review. This also translates into higher throughput for diagnostic reports and streamlined workflows that allow radiologists and cardiologists to focus their time and expertise where it’s most needed.
We are quickly expanding project Corazon to become a unified multimodal imaging platform (ultrasound, CT, cMRI, angio etc.). We want to standardize cardiac care across the U.S., democratizing earlier and even preventative diagnoses and interventions, by bringing under one roof this vision of a harmonized framework.
Q: Are there any infrastructure investments necessary to integrate its use into clinical care?
AK: Deploying AI models in a clinical environment may require institutions to make strategic investments across multiple infrastructure domains. First, are one or multiple cloud providers required for health systems to tap into dynamic graphic processing unit resources? Equally critical is a secure, HIPAA‑compliant system to manage, store and transfer medical images.
Also required is a rigorous security and governance framework: defined access roles, end‑to‑end encryption and continuous audit logging. Operational staffing is needed to monitor and resolve issues with vendor(s). With this framework, high levels of care — “action at a distance” — can be provided with the simplicity of an internet connection.
Q: Beyond imaging, your team is working on AI tools for scheduling and deidentifying patient health information. How do these tools address long standing pain points in hospital workflows and data management? What tools are on the horizon?
AK: We are first and foremost an engineering lab, meaning that we create foundational innovations and techniques, as much as practical focused applications. We have recently devised an AI-driven surgery-scheduling tool that can be extended to many other scheduling paradigms. Our system optimizes the allocation of limited resources while taking in stride the
dynamism inherent in clinical environments.
We also addressed a long-standing pain point for both research and health institutions: data
sharing in the presence of private health information. Upstream of our scheduling or
clinical imaging diagnostics, we saw the need for the pipeline to be expedited. In response, we made PixelGuard and NoteGuard meet these challenges, relieving two of hospitals’ most persistent bottlenecks.
Q: What should hospital and health systems leaders have been doing with AI — yesterday?
AK: The first necessary step is adoption of the infrastructure as outlined earlier. By cultivating a culture of data literacy and agile experimentation, they would be positioned to implement AI technologies at scale. With this infrastructure in place, hospital leaders would be poised to integrate AI into whichever layer of their operations they may choose.
Many AI technologies can be rolled out alongside routine care as silent clinical trials, where AI algorithms in parallel with routine clinical care contribute predictions or recommendations. This “shadow mode” can operate without influencing provider decisions, so their real‐world performance, safety and impact can be rigorously evaluated before live deployment. By comparing AI outputs to actual patient outcomes and clinician actions, these silent trials will help to identify biases, gaps and calibration needs, all while preserving patient safety.
Ideally, the adoption of intelligent tools would also foster cross‐functional teams. The Kline Engineering Lab regularly cross-pollinates ideas across our cardiac surgery, cardiology, radiology, engineering, computer science and data science departments. This dualism supports our co‐designing of AI tools that fit seamlessly into clinical needs and workflows.
Q: In terms of training in the incoming cardiovascular workforce, are training programs doing enough to prepare physicians to leverage AI?
AK: Most medical and surgical training programs underemphasize or underestimate the value of practical AI or data science skills. This leaves incoming physicians without hands-on experience and thus ill-equipped to appreciate how or why certain models or frameworks are better, how to evaluate them or leverage them for their own benefit.
To address this gap, fellowships should integrate specific AI modules, dedicated to the fundamentals of machine learning, statistical programming and data ethics. Ideally, the implementation of interdisciplinary rotations in engineering or data science labs would allow trainees and faculty collaborators to learn how to adopt an engineering mindset.
Programs that embed hands-on workshops or provide CME credits on emerging AI-driven research will ensure that both new and practicing physicians can stay current and can leverage AI responsibly and effectively. We have a CME AI conference being held in September. Our Bluhm Cardiovascular Institute’s Center for AI also has a MSAI-fellowship program that is open to Northwestern residents and medical trainees.