Dr. Evans, a radiologist, ends his day exhausted from the relentless volume of cases. Before he even finishes his first read of the morning, his worklist is already flooded with urgent studies, unread follow-ups, and new AI-generated alerts.
His schedule is packed – routine screenings, complex neuro cases, and high-priority findings all demanding his attention. The pressure to maintain accuracy while keeping up with demand leaves little room to breathe, let alone focus on the critical cases that require his expertise.
Radiologists across the country share this burden. The growing shortage of radiologists, combined with increasing imaging demands, is pushing radiology departments to their limits. As imaging remains a cornerstone of patient care, hospitals must find innovative ways to manage workloads without compromising diagnostic quality.
Artificial intelligence, particularly large language models, or LLMs, offers a compelling solution – not as a replacement for radiologists, but as a tool to enhance efficiency, reduce burnout, and improve clinical decision-making.
However, its deployment must be strategic, ensuring safe implementation, rigorous monitoring, and continuous validation.
A growing challenge
Some key statistics point to the scope of the challenges for radiology and radiologists:
Imaging volume is increasing by up to 5% annually, contributing to rising workload pressures on radiologists.
The U.S. may face a shortage of up to 42,000 radiologists by 2033, creating significant gaps in imaging services.
More than 45% of radiologists experience burnout, primarily due to increasing workload demands and staffing shortages.
Without innovative solutions, the gap between imaging demand and radiologist availability will continue to widen, affecting patient care and diagnostic efficiency.
LLMs can support radiologists
The sheer volume of imaging studies is unsustainable without smarter tools. LLMs and AI-driven automation offer relief by streamlining workflows, prioritizing critical cases, and reducing the manual burden of administrative tasks.
Radiologists spend a significant portion of their time summarizing patient charts, drafting reports, and reviewing clinical histories. LLMs can automate these repetitive tasks, allowing radiologists to focus on complex diagnostic work, including:
AI-Assisted Report Generation. LLMs can draft structured reports, reducing documentation time while ensuring consistency
Chart Summarization. AI can analyze prior imaging studies, clinical notes, and lab results to provide a concise case summary, aiding radiologists in decision-making
Safe implementation & post-monitoring
Despite the promise of AI, rushed or unvalidated deployment can introduce risks such as bias, workflow disruptions, and over-reliance on AI outputs. Implementation science must guide AI adoption, ensuring that models are continuously evaluated and monitored post-deployment.
There are several key areas that demand AI oversight:
Clinical Validation. AI models must be tested across diverse patient populations to ensure diagnostic accuracy and fairness
Bias Mitigation. AI should be monitored for unintended biases in prioritization, particularly in underrepresented demographics
Human-in-the-Loop Approach: Radiologists should always have final oversight, ensuring AI enhances – not dictates – clinical decisions
Post-Deployment Monitoring: AI performance must be continually tracked, with feedback loops allowing for updates and recalibrations
The real challenge isn’t just implementing AI – it’s ensuring that it delivers sustained, measurable improvements in radiology without unintended consequences.
Penn Medicine’s AInSights
Penn Medicine is at the forefront of AI-driven radiology advancements, with its AInSights initiative focused on the safe and effective deployment of AI in imaging.
Penn AInSights is an AI-powered radiology platform developed at Penn Medicine to enhance early disease detection and improve diagnostic efficiency. It automates image analysis, extracting quantitative data from scans and integrating AI-generated insights directly into radiology workflows.
The system has successfully processed thousands of imaging studies, reducing radiologist burden while ensuring key findings – such as liver steatosis and brain atrophy – are captured for early intervention.
Two recent peer-reviewed studies highlight the impact of this work:
One study details the development of a cloud-based system for automated AI image analysis and reporting, demonstrating significant efficiency gains and diagnostic value in radiology workflows (Chatterjee et al., 2024).
Another explores how AI-generated imaging traits can be integrated into common data elements (CDEs) to improve healthcare outcomes and workflow integration Mehdiratta et al.” (Mehdiratta et al., 2025).
What’s next with LLMs
Building on its success, Penn Medicine is now integrating large language models to further streamline radiology reporting. The goal is to automate the structuring of radiology report findings – such as detecting adrenal nodules – and trigger clinical decision support within EHR.
This next phase will improve reporting accuracy, reduce variability, and ensure that critical incidental findings prompt timely follow-up, optimizing both patient outcomes and resource utilization. Among the key objectives for AInSights:
Enhancing AI-Assisted Clinical Decision Support. AI tools are being developed to provide radiologists with deeper insights into imaging findings.
Post-Deployment Monitoring & Governance. AI models are rigorously evaluated to ensure real-world performance aligns with clinical expectations.
AI Integration with Workflow Efficiency. Efforts are underway to seamlessly incorporate AI into existing PACS, RIS, and EHR systems, reducing disruption to radiologist workflows.
AI FOMO? Take a strategic approach
The radiology shortage is real, but so is the pressure to deploy AI at lightning speed. With every new AI announcement, hospitals worry they are falling behind. However, implementing AI strategically – rather than reactively – is the key to long-term success.
For radiologists, AI isn’t just about efficiency, it’s about reclaiming time for complex cases, reducing burnout, and improving diagnostic accuracy.
But let’s be clear: Smart AI adoption beats rushed AI adoption every time. Instead of chasing trends, healthcare leaders must focus on implementation science, post-monitoring, and continuous refinement to ensure AI truly enhances radiology.
The future isn’t about who gets AI first – it’s about who gets it right.
Ameena Elahi is IS Application Manager at Penn Medicine, where she is responsible for project oversight for medical imaging applications, including research and artificial intelligence.
