Artificial intelligence is accelerating the detection of abnormalities on medical images, from tiny lung nodules to subtle lesions. In oncology especially, catching cancer early can be lifesaving. However, this heightened sensitivity comes with a hidden cost: more incidental findings. Studies estimate that roughly 15–30% of all adult imaging exams reveal at least one incidental finding—an abnormality found when looking for something else. [1] Most of these findings are benign, yet each triggers a clinical question: what next? Without careful management, more detection can paradoxically lead to more risk, burdening patients and healthcare systems with unnecessary procedures, anxiety, and potential harm.
When an MRI or CT scan uncovers an unexpected nodule or cyst, the instinct is often to investigate further “just in case.” Modern AI tools flag even tiny anomalies, and direct-to-consumer full-body MRI scans are now marketing the promise of finding hidden disease in healthy people. The result? A surge in incidentalomas—minor abnormalities that likely would never cause trouble. Whole-body MRI screening studies have shown that most follow-up biopsies prompted by incidental findings find no disease, with very few cancers detected, indicating substantial overtesting and overdiagnosis.[2]
In other words, many patients are “worried well,” chasing abnormalities that were never destined to harm them. Aggressively working up every incidental finding creates an illusion of benefit while actually causing physical and emotional harm from needless tests and treatments.
Of course, ignoring findings is also not the answer. Among the deluge of incidental alerts will be genuine early cancers or high-risk lesions that do need intervention. This is the other side of AI’s sword: false negatives. If an algorithm (or an overworked human following up dozens of AI-detected quirks) misses a true malignancy, that patient could fall through the cracks with dire consequences. Or, the more common scenario: an incidental finding pops up on a scan, but that follow-up recommendation stays tucked away in a radiologist’s report, never actioned or acted upon because of workflow gaps.
The challenge for radiology is to separate the signal from the noise—to act on the findings that matter and safely monitor or dismiss those that don’t—all while ensuring no critical finding is forgotten.
The Follow-Up Gap: Every Missed Follow-Up Is a Liability
For every abnormality identified, whether by AI or human, there is often a recommended follow-up: maybe a repeat imaging in 6 months, a referral to an oncologist, or an additional test to confirm a suspected diagnosis. These recommendations are the bridge between detection and outcome. Alarmingly, that bridge is often broken. Research shows that over 35% of recommended imaging follow-ups never get completed, and some analyses have found follow-up completion rates as low as 30% in certain settings. In other words, up to half of patients with actionable findings might not receive the next step in care.[3][4]
Missed or delayed follow-ups are a serious patient safety issue and a growing malpractice risk. Consider a scenario in which an AI flags a tiny lung nodule on a CT scan and the radiologist recommends a follow-up scan in 3 months.[5] If that follow-up never happens, and a year later the patient presents with advanced lung cancer, the liability for the institution is enormous. Without a system to track and ensure those recommendations are acted upon, it’s too easy for urgent findings to disappear into the ether.
Why do so many follow-ups fall through the cracks? The reasons are multifaceted: unstructured radiology reports that make it hard to flag actionable findings, communication breakdowns between radiologists and referring physicians, and patients who never schedule the advised test (or don’t even know they should). High-risk findings require high-reliability processes to ensure they don’t get lost in transition.
Turning AI Insights into Action: High-Reliability Follow-Up Care
To truly harness the promise of AI in radiology, hospitals must convert AI-driven insights into reliable action. It’s not enough for an algorithm to identify a potential cancer if the care team isn’t prepared to follow through every time. This is where principles of high-reliability organizations (HROs) come in—the idea that healthcare can emulate industries like aviation or nuclear power where mistakes are actively anticipated and prevented. For radiology follow-ups, that means building safety nets and redundancies so that no abnormal finding “falls through the cracks.”
In practice, that requires an approach that blends people, process, and technology:
- Structured Reporting and Communication: Encourage radiologists to make clear, unambiguous recommendations in their reports (e.g., specifying exact interval and modality for follow-up). Ambiguity or “optional” phrasing can lead to inaction.
- Automated Tracking Systems: Leverage AI and software to create automated processes for every actionable finding. Large Language Models (LLMs) can scan radiology reports for recommended follow-ups,” ensure they’re captured, translate them into workflows, and drive automation to coordinate care across patients and providers. This automation relieves clinical staff from the tedious tasks of manual tracking, ordering, and notifying patients. It frees them to focus on patient care. It also provides management with a dashboard view of open follow-ups, so they can intervene if a case is overdue.
- Patient Engagement and Navigation: Patients need to be empowered to manage their care. High-performing programs often employ nurse navigators or patient coordinators who contact patients with clear instructions for follow-up appointments. High-reliability systems can take it a step further, automating outreach to ensure the patient understands the recommendation and has the necessary support to complete it. The goal is to make following up as easy as possible for the patient, minimizing no-shows and drop-offs.
- Monitoring and Quality Improvement: Ultimately, institutions should consider follow-up completion as a key quality metric, alongside infection rates and readmission rates. By measuring follow-up rates and outcomes, teams can identify where breakdowns occur. Is the bottleneck in scheduling? Do certain departments have lower adherence? Regular audits and feedback to the care team create accountability. When clinicians see that follow-up compliance has climbed (or that missed follow-ups led to real harm), it reinforces the importance of these workflows.
The proliferation of AI in radiology is a boon for early cancer detection and diagnostic precision, but without a parallel investment in follow-up care processes, it could become a victim of its own success. More polyps, nodules, and lesions found mean little if they aren’t acted upon, or if they spur a wave of unnecessary interventions.
Radiology leaders have a chance to transform this potential risk into a patient safety triumph by building systems that ensure every important finding is not only detected but definitively managed. By weaving automation and accountability into follow-up workflows, we can fulfill AI’s promise to improve patient outcomes—finding the cancers that need treating, treating the ones that need it, and confidently monitoring those that don’t—all while never losing sight of a patient along the way.
Reference
[1] Davenport MS. Incidental Findings and Low-Value Care. AJR Am J Roentgenol. 2023 Jul;221(1):117-123. doi: 10.2214/AJR.22.28926. Epub 2023 Jan 11. PMID: 36629303.
[2] Richter A, Sierocinski E, Singer S, Bülow R, Hackmann C, Chenot JF, Schmidt CO. The effects of incidental findings from whole-body MRI on the frequency of biopsies and detected malignancies or benign conditions in a general population cohort study. Eur J Epidemiol. 2020 Oct;35(10):925-935. doi: 10.1007/s10654-020-00679-4. Epub 2020 Aug 29. PMID: 32860149; PMCID: PMC7524843.
[3] Mabotuwana T, Hall CS, Tieder J, Gunn ML. Improving Quality of Follow-Up Imaging Recommendations in Radiology. AMIA Annu Symp Proc. 2018 Apr 16;2017:1196-1204. PMID: 29854188; PMCID: PMC5977608.
[4] Stempniak M. Patients frequently fail to obtain follow-up imaging. Could radiologist-referrer disagreements be to blame? Radiology Business (Practice Management). November 18, 2024. Online. Last accessed on May 31, 2025.
[5] Ridley EL.How can radiologists ensure report recommendations are acted on?Imaging Informatics Enterprise Imaging Jul 24, 2022. Online. Last accesses on May 31, 2025.
Featured image © 2016 – 2025 Hitesh Choudhary. Under Unsplash licenses. Used with permission.
DOi
