Modern medicine has never had greater access to imaging, patient data, computational intelligence, and tools than it does today. And yet, clinical decision-making at the point of care hasn't improved at the same pace. The conversation around clinical AI tends to default to patient dashboards or autonomous diagnosis while overlooking the fundamental challenge of how technology should actually integrate with the workflows of physicians responsible for life-altering decisions.
Prithvinath Garigapuram, CEO and Co-Founder of CARA Systems Inc., an NYU spinout building AI-assisted decision-support tools for neurovascular care, believes that the field has been solving the wrong problem. In his view, the challenge isn't merely developing AI systems capable of generating clinical predictions, but building intelligent clinical co-pilots that can contextualize complex patient-specific data and translate it into meaningful decision support for physicians in real-time.
The Real Problem Isn't Data. It's Decision-Making
Inside a neurovascular clinic, the current constraint doesn't come down to a lack of information. A JAMA analysis of seven U.S. integrated health systems and Ontario found that, among older U.S. adults, annual CT use roughly doubled between 2000 and 2016, from 204 to 428 exams per 1,000 person-years, while MRI use more than doubled, from 62 to 139 per 1,000. To put it plainly, electronic health records hold more longitudinal data than any clinician can read.
The constraint is what happens when it comes to reviewing a scan and recommending a practical course of action. Healthcare decision-making, as Garigapuram describes it, eventually comes down to what a physician interprets based on a patient's symptoms, clinical variables, and overall presentation, regardless of the sophistication of the technology involved.
The challenge ultimately lies in the quality and context of the information available to the physician at the moment a decision must be made. Evaluating a patient often requires synthesizing multiple interconnected data streams, imaging findings, clinical history, anatomical variability, physiological indicators, laboratory data, and individualized risk factors, all of which influence one another in ways that no single scan or isolated metric can fully capture.
In practice, much of this integrated assessment still occurs cognitively within the physician's mind, often under significant time pressure and with incomplete patient-specific evidence. As a result, clinicians are frequently forced to rely on their experience and fragmented information when making highly individualized care decisions.
The way Garigapuram sees it, much of clinical AI to date has been mostly centered around patient dashboards, single-modality prediction models, or autonomous-diagnostic systems that operate adjacent to rather than within the physician's workflows. The limitation of this approach becomes apparent in clinical settings. A model trained to flag and identify aneurysm risk from a CT angiogram alone, for example, may recognize imaging patterns with high statistical confidence, yet remain blind to the broader clinical context a physician evaluates, like a patient's hypertension history, family history, symptom presentation, prior imaging, and other patient-specific factors that fundamentally shape risk interpretation.
The result is often a system capable of producing highly confident outputs from a clinically incomplete picture, leaving physicians to reconcile the recommendations with the contextual information the model never incorporated. While these systems may demonstrate strong performance in controlled validation environments, many struggle to integrate meaningfully into real-world clinical practice.
What Garigapuram Believes Physicians Actually Need
Garigapuram's view, drawing from years of collaboration with neurospecialists at NYU Langone during his work in medical robotics at NYU, is that technical capability alone isn't enough to meaningfully transform healthcare. As he puts it, "Developing meaningful healthcare technology requires far more than technical innovation alone. It demands empathy, close collaboration with clinicians, and a deep understanding of patient care realities, clinical workflows, and the human impact behind medical decision-making."
Designing for that reality means recognizing that clinical decisions emerge from the simultaneous evaluation of imaging findings, patient history, anatomical variation, longitudinal progression, and individualized risk factors. Any system that fails to account for this interconnected reasoning disconnects from actual clinical practice, regardless of how well it performs in controlled testing environments. Garigapuram also highlights the importance of embracing the inherently regulated, long-cycle nature of healthcare innovation, where clinical collaboration, pilot validation, and physician trust are essential to establishing real-world credibility and adoption.
Clinical AI built in isolation from clinicians tends to produce systems and tools that look impressive in research settings but fail to influence clinical practice at scale. The diagnosis underneath Garigapuram's thesis is that the core gap isn't technical capability but the absence of systems capable of integrating multimodal clinical information into a coherent, physician-centered layer of decision support.
What physicians actually need, in Garigapuram's framing, is the opposite of a single-modality black box: a tool capable of synthesizing the same inputs they evaluate and presenting them as a contextualized patient-specific integrated picture to support real-time decision making.
The Co-Pilot Model: Decision Support Done Properly
The alternative Garigapuram advances is a decision-support model: AI that unifies the data streams clinicians already use, surfaces patient-specific insights, and integrates directly into existing workflows. At CARA Systems, that framework takes the form of a clinical decision intelligence engine built for neurovascular care. The platform is designed to process patient-specific data by combining advanced medical image analysis, computational biomechanics, hemodynamic modeling, and predictive analytics into clinically interpretable outputs that physicians can meaningfully act upon.
Much of Garigapuram's work has focused on developing workflows capable of transforming non-invasive medical imaging into patient-specific anatomical characterization, rupture-risk assessment, and computational blood-flow analysis. He is also a published inventor on a U.S. patent application related to intracranial aneurysm assessment. Importantly, the system is not designed to function as a black-box prediction engine.
The emphasis is on interpretability and clinical transparency. Rather than presenting physicians with a single isolated risk score, the platform is intended to surface the anatomy, flow dynamics, structural relationships, and underlying reasoning contributing to an assessment, allowing clinicians to evaluate the broader clinical context alongside the model's insights.
The implications of that approach are practical and clinically significant. Conventional aneurysm assessment pathways often escalate patients toward invasive angiography procedures when non-invasive imaging alone cannot sufficiently resolve uncertainty. Those procedures carry their own procedural risks, costs, and resource burdens. CARA's engine is built explicitly as a non-invasive triage layer, with the goal of surfacing enough patient-specific insight upstream to keep lower-risk patients out of the cath lab and flag the patients who genuinely need it sooner.
That distinction reflects the broader difference between a clinical co-pilot and a purely autonomous diagnostic system. The value is not necessarily in producing a different verdict, but in enabling a more informed, contextualized, and clinically grounded pathway to reaching one.
"Our goal at CARA is to build tools that enhance clinical insight, reduce uncertainty, and support better decision-making at the point of care," he says. In many ways, the principle extends beyond neurovascular care. The broader opportunity in healthcare AI, to him, isn't centered around developing models anchored by performance and output, but designing decision-support systems and frameworks capable of contextualizing complex patient-specific information in a way that meaningfully supports clinical decisions and integrate naturally into physician workflows.
His Vision for the Future of This Framework
Garigapuram positions neurovascular disease as an early proving ground for a broader clinical intelligence framework. In his view, the same foundational principles driving the work at CARA Systems Inc. can extend across multiple areas of medicine where decision-making is complex, high-stakes, and time-sensitive, including oncology, cardiovascular disease, and advanced surgical planning.
The broader objective is a shift from reactive to a more predictive, personalized, and proactive model of care. Garigapuram believes the future of healthcare will increasingly need systems physicians can use to synthesize large volumes of patient-specific information in real-time, enabling earlier interventions, more precise risk stratification, and better-informed treatment pathways.
Achieving this transition, however, requires more than incremental improvements in AI performance. It demands a fundamental rethinking of how healthcare technology is designed and integrated into clinical practice. He also emphasizes that progress in this space depends heavily on multidisciplinary collaboration between clinicians, engineers, researchers, and healthcare institutions working around shared clinical problems and objectives. Many of the most important challenges in medicine, he argues, cannot be solved through innovation alone without a deep understanding of clinical realities, workflow constraints, and patient care dynamics.
For the broader category of clinical AI, the implication is that adoption will follow technologies designed around the physician's workflow, contextual decision support, and measurable patient impact. As Garigapuram himself puts it, "Healthcare technology and innovation become truly meaningful when it helps physicians make faster, more informed decisions that can directly improve patient outcomes."
The position Prithvinath Garigapuram is advancing puts clinical AI as a co-pilot designed for real workflows and evaluated by its ability to improve care delivery and patient outcomes. It is a direction many believe the field is likely to converge on as the limitations of purely autonomous-diagnosis framing become more apparent.
