May/June 2026 Issue
AI Insights: Balancing the Load
By Khan Siddiqui, MD
Radiology Today
Vol. 27 No. 3 P. 8
Why Image Detection Isn’t Radiology’s Biggest AI Problem
Radiology is facing growing capacity pressures. Because of scanning technology innovation and longer life expectancies, the sheer volume of medical imaging data requiring review is now growing three times faster than the number of practicing radiologists. Over time, this imbalance is contributing to rising workload fatigue for radiologists, making it harder to plan for the future while meeting patient needs.
As imaging demand increases, imaging AI that can expedite scan interpretation is often discussed as part of the response. However, reviewing imaging isn’t the true workflow bottleneck, as radiologists only need 250 milliseconds to spot findings on an image. Rather than looking to AI to replace radiologists’ “findings detection” task, imaging AI can help simplify the administrative aspects of their workflow by predrafting reports, importing measurements directly into reports, and helping prioritize the review of high-risk cases.
AI-supported workflows could alleviate some of the chronic strain within the profession. The 2025 Physician Burnout & Depression Report by Medscape states that 51% of radiologists report feeling burned out. Another industry study found that 67% of practices are understaffed, a factor that often contributes to burnout.
These numbers reflect the reality that scanning technology has greatly increased the volume of images radiologists review. One medical center found that between 1999 and 2010, the total number of images generated jumped from 9.2 million to 94.2 million per year. Their study also found that, to keep up with the new volume of data, a radiologist would have to interpret one image every three to four seconds for the entire duration of an eight-hour shift.
Radiologists bring years of training and judgment to diagnosis, which is central to their enjoyment of their work and patient care. By reducing the documentation and administrative tasks of the job, we can help ease the demands of the profession and support an attractive, sustainable profession that forms a pillar in state-of-the-art patient care.
Many imaging platforms look to add new capabilities for specialized tools to the radiology workflow. However, radiologists review imaging inside defined PACS, and opening a different portal or software costs time and interrupts workflows. To be useful, AI tools need to be embedded within PACS and feed their results directly into reporting templates, a feature most new platforms don’t have.
Adding dozens of specialized AI tools to the radiologist’s workflow can also generate alerts that are more distracting than helpful. In terms of Daniel Kahneman’s two systems of thinking, this transition likely moves image interpretation from rapid, System 1 intuition toward a more effortful and deliberate System 2 process, further increasing cognitive load and contributing to the pervasive mental fatigue now straining the profession. In some cases, AI tools may flag clinically irrelevant data or generate a false positive, which may make the radiologist legally obligated to explain in their report why they disagree with the tool, which takes more time.
A Human-Technology Partnership
Most radiologists now use speech-to-text dictation software to record their findings while scrolling through images. However, the result is often a collection of observations, and the radiologist has to spend time turning these observations into a single report. This is where studies have already found that AI can be genuinely helpful, increasing the speed of report generation by more than 15%, without altering clinical accuracy or textual quality.
Typos and misdictation in reports can lead to serious confusion and distress for patients and increase the chances that the patient will reach out for clarification, resulting in more work for the physician. AI tools, especially large language models (LLMs), are well-suited to finding communication errors and discrepancies. One review also found that they are better than most physicians at responding with empathy, while another discovered that including an LLM-generated, patient-friendly version of the report can mitigate patient anxiety and confusion.
Largely because of the prevalence of LLMs, people tend to generalize all AI tech as interpretive, but some of the biggest opportunities to improve the field of radiology have nothing to do with interpreting scans. AI can help night shift radiologists manage workload more effectively by automating worklist sorting, surfacing case age and order type, and reducing administrative coordination like routing cases or flagging delays beyond facility thresholds. Since a single radiologist may cover all studies overnight, efficient queue management is critical. By streamlining logistics, AI can reduce strain on local teams, decrease reliance on teleradiology, and shorten time from image acquisition to physician review, even during the day.
AI tools can also review physician orders and ensure that the scanner is set up with the correct settings before the patient even enters the room. Others can detect if patients move during scans—prompting the technologist to retake it while the patient is still present—and feed exact measurement data directly into the draft report.
Expecting imaging AI to do the primary diagnostic work of radiologists is asking technology to solve the wrong problem. What the field really needs are tools that meld seamlessly with existing software, reduce time spent on tasks beyond detection, and provide a safety net that relieves administrative stress, reduces mental fatigue, and increases patient understanding. This new, AI-supported workflow is one that returns autonomy and professional fulfillment to a specialty that has long been buried under the weight of its own data.
— Khan Siddiqui, MD, is the cofounder, chairman, and CEO of HOPPR.
