The development of AI has impacted almost every industry,
but the medical field, and in particular nuclear medicine, has been quick to
find ways to adopt AI.
In March of 2026, the SNMMI AI Task Force held a summit to
ensure that nuclear medicine stays on the cutting-edge of AI technology.
Experts gathered to address the challenges and opportunities alike, seeking to
identify meaningful ways in which AI can help nuclear medicine professionals in
their daily workflows.
For a recap of the AI Summit, we talked with AI Task Force Chair Abhinav Ja, and Vice Chair Tyler Bradshaw.
1. The theme of the summit was Clinical Adoption of AI. What was the
central goal of the summit, and why is this moment particularly important for
focusing on clinical adoption of AI in nuclear medicine?
Thank you for the opportunity to discuss the summit. Our
goal was to advance the conversation on artificial intelligence in nuclear
medicine from promise to prctice by bringing together stakeholders from across
the nuclear medicine ecosystem to focus on a shared challenge: how do we
translate advances in AI into clinical practice?
Our field is seeing rapid advances in AI development,
leading to the pressing question of how these methods can be translated into
tools that reliably improve clinical care. We are at an inflection point, as
many AI approaches are showing promise, yet their adoption in routine clinical
workflows remains limited. This gap reflects broader issues related to
validation, integration, and trust.
It is important for us to carefully consider how to
translate these advances into clinically meaningful improvements in patient
care. This was the central focus of the summit.
2. How was the summit intentionally structured to foster meaningful
dialogue and progress toward clinical adoption?
To advance clinical adoption of AI solutions, our community
needs to navigate several key challenges, including clinical validation,
regulatory pathways, implementation in practice, development of data
ecosystems, academic industry partnerships, and economic sustainability,
including reimbursement models.
To address these challenges, our vision for the summit was
to serve as a platform for broad community engagement, learning from key
opinion leaders and incorporating perspectives across stakeholders. In line
with this vision, the summit was structured into multiple sessions focused on
these areas, allowing for more targeted discussion and reflection. Sessions
included invited talks from key opinion leaders followed by panel discussions
to further explore these themes.
In addition, the summit included breakout sessions where
attendees discussed topics such as implementation and validation of AI models,
and academic industry partnerships. Overall, these discussions helped generate
important insights and actionable perspectives on how to strengthen the
adoption of AI in nuclear medicine.
3. What do you see as the most important challenges currently limiting
clinical adoption of AI in nuclear medicine, and how can the field begin to
address them?
A key theme that emerged from the summit was that clinical
adoption of AI progresses along a continuum from clinical validation to
improvements in clinical efficiency, and ultimately to demonstrating clinical
utility. While there are multiple challenges across this continuum, there are a
couple I would like to highlight.
The first is ensuring that AI solutions are developed with
the end user and clinical need in mind. AI solutions that are not driven by
clearly defined clinical tasks can make it difficult for clinicians to see how
these tools fit into their workflows or meaningfully impact patient care.
Closely related to this is the need for rigorous and
clinically meaningful approaches to validation. Even when AI methods show
strong performance on conventional metrics, these do not always translate into
improved clinical decision-making or outcomes. This creates a gap between
development and adoption and contributes to hesitancy among clinicians.
Addressing these challenges will require a more deliberate
focus on defining clinical use cases up front and aligning both development and
evaluation with those use cases. The summit highlighted the importance of
aligning development and validation with clinical needs, as well as the role of
emerging approaches including in silico and data-driven methods in enabling
more robust and scalable validation. It also emphasized the importance of
bringing together diverse stakeholders to ensure that AI solutions are both
technically sound and clinically relevant.
4. How should the field think about validating AI systems in a way that
ensures they deliver meaningful clinical value?
To ensure that AI systems deliver meaningful clinical value,
their evaluation needs to be grounded in clearly defined clinical tasks. This
includes assessing performance in ways that reflect how the system will be used
in practice, evaluating generalizability across settings, and accounting for
potential sources of drift between training and real-world data. Importantly,
evaluation studies should yield quantitative claims that are tied to specific
clinical tasks and supported by transparent study design and appropriate
performance metrics.
In this context, there has been increasing emphasis within
the field on structured approaches to evaluation. For example, The Journal
of Nuclear Medicine has recommended that evaluation studies of AI
algorithms provide a claim that clearly defines the clinical task, describes
the study design, and reports quantitative figures of merit that reflect
performance for that task.
More broadly, frameworks such as the RELAINCE guidelines
provide a way to perform evaluation across the lifecycle of an AI system, from
proof-of-concept to post-deployment, and offer best practices for these
different classes of evaluation.
Finally, it is important that the evaluation be
multidisciplinary, with active involvement from clinicians, to ensure that the
results are both technically sound and clinically meaningful.
5. Based on the discussions at the summit, where is AI currently best
positioned to make a meaningful impact in nuclear medicine?
There are multiple areas where AI is well-positioned to make
a meaningful impact, particularly in supporting healthcare providers and
improving patient care. In addition to improvements in areas such as image
reconstruction and quantitative analysis, one area that generated significant
interest at the summit was the use of AI in theranostic applications.
AI has the potential to contribute across several stages of
the theranostic pathway, including identifying patients most likely to benefit
from treatment, developing optimized, practical, and clinically implementable
imaging and dosimetry protocols, and predicting therapy response and other
clinical outcomes, including through mechanisms such as digital twins. These
are areas where improved decision-making and quantitative accuracy can directly
influence patient outcomes.
The session on AI in theranostics highlighted a range of
advances in this space and underscored both the opportunities and the important
questions that remain. Overall, this area represents a compelling example of
how AI, when aligned with well-defined clinical needs, can have a meaningful
impact on nuclear medicine practice.