This insight was featured in the January 14th, 2026 edition of the AI Catalyst Pulse.
Approximately 77% of FDA-approved AI-enabled devices are focused on radiology and imaging, making it the clear frontrunner in AI-powered clinical decision support (CDS). Radiology’s digital first workflow allows AI to integrate naturally across the imaging lifecycle, unlocking opportunity for greater accuracy, efficiency, and improved clinical outcomes.
Yet adoption has not been without friction. Clinicians face rising exam volumes, increasing exam complexity driven by an aging population, and growing workforce strain as burnout and attrition rates climbed from 1.9% to 3% post-pandemic. Radiologists exemplify the tension that we see across healthcare more broadly: the gap between healthcare supply and demand is widening. At the same time, hesitancy remains around AI tools that influence clinical decisions rather than simply reducing administrative burden.
Radiology’s experience offers a preview of what’s coming for AI-enabled CDS across healthcare. Health systems that apply these lessons will be best positioned to address the growing imbalance between clinical demand and available capacity.
Our latest briefing outlines four core principles leaders must embrace to maximize the impact of AI-powered CDS:
1. Make AI invisible through workflow integration. When deploying AI tools for ambient documentation, success often depends on training clinicians to optimize how they use the technology. AI-powered clinical decision support, however, requires the opposite approach: the technology must adapt to the clinician, not the other way around. Sutter Health demonstrates this approach through its use of CDS algorithms for imaging challenges such as lung nodules and cancer detection, in partnership with Ferrum Health. The lung nodule tool succeeds because it is non-intrusive, running post-dictation as a safety net, adding clinical value without disrupting how radiologists work.
2. To boost adoption, position the human as the decider by involving them early, framing AI as supportive, and embedding safeguards that preserve autonomy. Sutter Health exemplified this approach by engaging clinical experts, including chest specialists and thoracic imaging fellows, to test and validate its lung nodule detection tools. Early involvement of trusted peers built confidence and accelerated adoption among radiologists.
Duke Health’s Surgical Artificial Intelligence and Innovation Laboratory (SAIIL) applied the same principle in developing AI for live surgical video analysis. Because surgeons value autonomy in the operating room, real-time AI alerts can feel disruptive. To respect surgeons’ autonomy and gain surgeon buy-in, Duke framed the AI as a legal protection rather than liability. Surgeons were engaged from the outset as expert contributors, with plans for customizable user interfaces to further reinforce clinician control.
Adventist Health similarly reinforced human authority with its KATE AI triage tool. Nurses retain final decision-making power by choosing whether to accept or reject AI recommendations, while also providing direct feedback - ensuring clinicians remain in control and enabling continuous real-time improvement.
3. Let AI do the work that ‘wastes’ expertise. UMass Memorial Health adopted the AEYE-DS tool for diabetic retinopathy screening and, during its pilot phase, reduced screenings by 75%. While the tool has since been discontinued, the time savings enabled physicians to focus more on direct patient care, patient education, and timely communication of results - driving improved outcomes and stronger adherence to clinical guidance.
4. Reframe value expectations (for AI CDS tools) from cost-savings to clinical confidence and safety. In imaging, AI can be integrated at multiple points in the workflow, but its impact ultimately depends on what happens after the report is completed. Outcomes are shaped by downstream clinical decisions, timeliness of follow-up, and patient access to care. In most cases, measurable ROI from efficiency gains should not be expected immediately. Instead, the value of AI-powered CDS lies in building clinical and executive confidence.
Adventist Health’s KATE AI solution demonstrates this approach by supporting frontline nurses amid declining levels of clinical experience. Rather than focusing on near-term cost savings, the value of KATE AI lies in strengthening clinical judgment and giving both nurses and executives greater confidence in care delivery.
Radiology’s experience with AI-powered CDS has shown that the true value lies not in replacing clinicians, but in strengthening their confidence and capacity to deliver timely, high-quality care. As AI-powered CDS expands across specialties, health systems that apply these lessons will be best positioned to realize meaningful clinical impact.
Read our complete research brief for more details on these four key lessons and case studies from successful implementation.