Making Follow-ups Count
For the past three years, the ACR Learning Network’s ImPower Program has been helping radiology practices and departments answer a simple question: How can we do better?
Read more
From the Chair of the Commission on Economics
As artificial intelligence (AI) tools — and other advanced computer analytics — continue to gain traction in radiology, it is important to understand relevant rules governing their coding and billing. This is particularly true now that some codes for AI are being reimbursed by Medicare and other payors. This column outlines foundational billing principles, reviews some AI taxonomy and highlights common coding pitfalls.
To accommodate the expected rapid expansion of AI tools, the AMA’s Current Procedural Terminology (CPT®) Editorial Panel, with input from ACR, has created a family of quantitative image analysis codes. Many of these codes are structured in pairs: one for analysis performed with a concurrent diagnostic examination, and one as a stand-alone service analyzing a prior dataset. These are currently Category III codes (temporary codes for tracking emerging technologies), but they are being reimbursed in some settings.
Examples include:
There are other, more narrowly focused, AI CPT codes (e.g., cardiac fractional flow reserve) that have relatively straightforward coding and application, but these broader codes raise questions about appropriate use. Is Hounsfield measurement of the liver a billable quantitative tissue characterization? What about automated volume measurement? Can these be routinely added to every diagnostic examination?
At the heart of Medicare billing, and integral to many private insurance contracts, is the requirement that services be medically necessary. This mandate stems from the Social Security Act (Section 1862(a)(1)(A)), which prohibits Medicare from paying for items or services that are “not reasonable and necessary for the diagnosis or treatment of illness or injury.”
That requirement applies not only to the base imaging study but also to any adjunctive AI services. Documentation must support medical necessity for each billed component.
AI services must also not duplicate work already captured in the base imaging CPT code. To be separately billable, the analysis must contribute independent clinical value beyond what a radiologist would typically provide and that value must be reflected in the radiology report. For example, measurement of pulmonary nodules and liver Hounsfield units are services already included in the base imaging interpretation codes. AI automation of those tasks does not qualify for separate reimbursement.
If an algorithm performs both assistive and augmentative functions, the augmentative component must be independently medically necessary to justify billing an augmentative CPT code.
Proper coding is always important but especially so here, as the addition of AI codes may subject patients to out-of-pocket coinsurance expenses they were not anticipating.
Appendix S of the CPT® 2025 Codebook introduces a taxonomy of AI functionality. Two categories are particularly relevant:
For example, consider a hypothetical advanced algorithm that detects focal liver lesions on an abdominal CT (medically necessary but duplicative) and quantitatively calculates the probability of the patient having blue eyes (augmentative and interesting but not medically necessary). It is not appropriate to utilize an augmentative tissue characterization code for this service despite it being both medically necessary in part and augmentative in part. The portion of the service described by the CPT code (augmentative tissue characterization in this example) must be both medically necessary and non-duplicative.
The ACR Commission on Economics has been actively working to develop a coding framework that can evolve to ensure appropriate reimbursement for AI tools as they become more commonly used. Understanding the coding and billing rules underlying that framework is important for regulatory compliance and to ensure accurate Category III coding data, which can support future Category I applications and ultimately broader reimbursement.
Making Follow-ups Count
For the past three years, the ACR Learning Network’s ImPower Program has been helping radiology practices and departments answer a simple question: How can we do better?
Read moreIndependent Dispute Resolution Under the No Surprises Act — The Basics
What are the consequences of the federal IDR process for plans, providers and patients?
Read moreAMA Still Needs Help with Survey on Indirect Expenses
Radiology practitioners have been mostly unresponsive, not realizing the information is being gathered in the hopes CMS will use it to calculate reimbursement.
Read more