top of page

AI Integration and FMV Radiologist Compensation

3 hours ago
7 min read
Lin Hsu, CVA, Manager at BuckheadFMV

AI is making radiologists faster, but it is not making any single read less consequential. If anything, every additional study a radiologist signs carries the same professional and legal exposure as the previous one. That tension sits at the center of a question health systems, radiology groups and valuation professionals will increasingly face: when AI raises productivity, what is fair market value (FMV) compensation for the radiologist doing the work?

The question came into focus during a recent industry webinar on what panelists called an “AI-native” radiology operating model. Panelists, including the leader of a radiology technology company and its affiliated teleradiology group, a health system chief medical officer, and the group’s radiology chair, described a platform in which AI can generate a preliminary read for each study, with the radiologist providing the final interpretation. AI agents also handle much of the clerical work around the read, and radiologists work on demand and are paid per work RVU (wRVU).

The panel warned against a practice some large groups have begun adopting: lowering radiologists’ pay because AI helps them read more. A radiologist who reads more studies accepts liability for more patients, and that risk does not shrink because software assisted with the reading.

How AI in radiology is evolving

Radiology remains the center of gravity for clinical AI. As of March 30, 2026, the FDA had cleared 1,524 AI-enabled algorithms, of which 1,163 (about 76%) were radiology products (Radiology Business). Early tools focused on narrow detection and triage, such as flagging a likely intracranial hemorrhage. The model described in the webinar uses thousands of software agents across the entire workflow, including:

  • Checking that studies arrive complete, with the right sequences, tech worksheets, and ordering comments

  • Building “smart” worklists that weigh service-level agreements, urgency, and subspecialty

  • Auto-populating report templates tailored to each referring physician’s preferences

  • Flagging findings, priors, or billing elements a report has missed before it is signed

  • Running silent “shadow” reads that send discrepancies to peer review

The panel estimated that radiologists historically spent roughly 40% of their time on clerical tasks. Remove that, and the same radiologist can sign meaningfully more studies per hour. In this case, the group reported an 18% increase in productivity from its worklist changes alone.

AI is not signing reports. AI-drafted preliminary reports are optional; radiologists can reject AI findings, and the radiologist remains the physician of record on every study.

The compensation question

When productivity rises, organizations revisit pay. Most radiology arrangements tie compensation to wRVUs, so if AI enables a radiologist to produce 20% more wRVUs per shift, pay under a per-wRVU model would rise accordingly.

Some organizations see that as a windfall created by technology they paid for, not by the physician. In response, some large radiology groups are lowering the per-wRVU rate after implementation to capture more of the gain.

The webinar’s teleradiology group pays a flat base rate per wRVU, with premiums for peak hours, weekends, holidays, and surge periods, and does not reduce the rate based on efficiencies created by its platform. Its position is grounded in one key point: more reads mean more liability, and the radiologist, not the software, bears that responsibility.

Liability scales with volume, not with effort

Malpractice exposure in radiology scales with the number of studies a physician signs off on. One teleradiology chief medical officer’s analysis of claims data found that an average radiologist can expect at least one malpractice claim for roughly every 100,000 studies read, or about once every seven years (AuntMinnie). A radiologist who reads 20% more studies reaches that threshold sooner, roughly every six years instead of seven.

AI does not move that risk to the vendor. U.S. malpractice doctrine generally still assigns liability for negligence to the physician when AI is involved (AJMC). AI may even increase exposure: a 2026 mock-jury study in Nature Health found jurors were more likely to fault a radiologist who disagreed with a correct AI flag after a single review than one who first read the scan independently (AuntMinnie).

This creates a practical tension. Safeguards that reduce liability, such as independent reads before AI review, comparison with priors, and complete documentation, take time. Cutting per-wRVU rates pushes radiologists to read even faster to protect income, eroding those safeguards. As the radiology chair put it, reporting fast and reporting well are different skills, and a productivity model has to reward both.

What this means for FMV compensation

For arrangements involving hospitals and referral sources, compensation must be consistent with fair market value and commercially reasonable under the Stark Law and Anti-Kickback Statute. Productivity pay based on a physician’s personally performed services is generally permissible, so higher wRVU output from AI-assisted reading can support higher total compensation. The harder questions are about the rate and the inputs.

How AI productivity gains should be treated depends on how the radiologists are engaged. BFMV performs FMV work on both of the arrangements below, and the questions differ.

Private groups receiving hospital support. Many hospitals contract with independent radiology groups for professional coverage and pay a subsidy or stipend when the group’s professional collections do not cover the cost of providing the required coverage. The hospital pays the group, not individual radiologists, and the group sets its own physician compensation. AI-driven productivity can increase collections or allow the same coverage with fewer radiologist hours, which may reduce the financial need the support was meant to address. A hospital may reasonably revisit the support amount, but the analysis should consider who funded the technology, the coverage requirements (such as 24/7 availability, turnaround times, and subspecialty reads), local recruiting and retention conditions, and the liability the group’s radiologists continue to carry. Efficiency gains the group invested to achieve are not automatically the hospital’s to recapture.

Employed radiologists. Hospital- or health system-employed radiologists are typically paid a base salary, wRVU-based productivity pay, or a combination, benchmarked against compensation surveys. The employer usually supplies the AI tools, workstations, malpractice coverage, and benefits, so under a per-wRVU model, productivity gains flow directly to the individual radiologist. Here the question is whether the per-wRVU rate and total compensation remain consistent with FMV as output rises, and whether a rate reduction is supportable given the work and liability the radiologist still carries. Employer-provided technology and coverage are part of that analysis.

Across both models, we see eight areas where AI changes the analysis.

1. Per-wRVU rates should reflect the work, including the stress. Work RVUs reflect the time, skill, judgment, and stress involved in an interpretation. AI may speed things up, but the responsibility and pressure to get it right remain with the radiologist. Lowering the rate solely because reads got faster may understate what the radiologist provides.

2. Survey benchmarks lag the technology. Surveys largely predate the widespread adoption of AI and often show compensation per wRVU flattening at high productivity levels. Applied mechanically to an AI-enabled radiologist at very high percentiles, they can produce unsupportable total compensation. High output should be tested against quality metrics, call obligations, and the market for comparable radiologists.

3. Who pays for the AI and other technology matters. AI licenses, cloud PACS, worklist software, reading workstations, and connectivity carry real costs. When a health system supplies them, that support has value and may affect what compensation is consistent with FMV. When the group or radiologist pays, those costs need to be factored into an FMV analysis. Either way, the funding party should be identified and documented.

4. EHR access should be scoped. Radiologists need access to the EHR to review clinical history and prior results. Access limited to the contracted services is generally part of the arrangement. Still, access or equipment that extends beyond them, such as use for the group’s other patients, can be remuneration and may need to fit the Stark exception and Anti-Kickback safe harbor for EHR items and services, or be paid for at fair market value.

5. Malpractice coverage is a real cost of the read. Premiums may rise with study volume. If the radiologist or group pays them, that cost belongs in the analysis; hospital-provided coverage shifts the calculus. Tail coverage, policy limits, and whether policies clearly cover AI-assisted reads should also be addressed.

6. Benefits change the comparison. On-demand radiologists are often independent contractors who fund their own benefits and self-employment taxes, while surveys generally report cash compensation for employed physicians. Comparing the two directly can understate a contractor’s appropriate rate. For employed radiologists, the benefits package should be valued alongside cash pay.

7. State practice limitations shape the arrangement. Teleradiologists generally must be licensed in the patient’s location and credentialed at each facility, and someone bears those costs. Some states’ corporate practice of medicine and fee-splitting rules limit how non-physician entities can employ radiologists or share in professional fees, and a growing number of states regulate the use of AI in healthcare. These rules affect who may provide the service and what rate is reasonable.

8. Quality guardrails support the rate. Documented peer review, discrepancy tracking, and AI-assisted quality checks show that higher volume is not coming at the expense of care, strengthening the commercial reasonableness of paying for it.

The FMV analysis should account for both the efficiency AI provides and the liability the radiologist continues to absorb.

The bottom line

AI is changing how radiologists work, but not who is accountable for the read. Organizations tempted to capture AI efficiency gains by cutting per-wRVU rates or group stipends should weigh that move against the liability radiologists continue to carry and the quality safeguards that take time to maintain. A defensible FMV analysis recognizes both.

BFMV can help health systems and radiology groups evaluate whether financial support for radiology groups and compensation for employed radiologists in AI-enabled settings are consistent with fair market value and commercially reasonable.

This article is for general informational purposes and is not legal advice; specific arrangements should be reviewed with qualified healthcare counsel.

Sources

Featured Posts
Recent Posts
Search By Tags
Archive
bottom of page