Regulatory & Compliance
AI in Radiology: Coding, Compliance, and the Road Ahead
July 09, 2026 - Melody W. Mulaik, CRA, FAHRA
Clinical artificial intelligence is no longer a distant concept in radiology. It is here and it is being coded and billed for today. For radiology administrators, revenue cycle professionals, and compliance teams, that shift from theoretical to operational is significant. What began as pilot programs and CPT® Category III tracking codes has begun its transition into Category I territory, bringing with it the full weight of documentation requirements, coverage policies, and billing rules that govern every other imaging service we report.
At the same time, the regulatory framework surrounding AI in radiology remains a work in progress. The Centers for Medicare and Medicaid Services (CMS) is watching, payers are developing policies, and organizations that move forward without a clear understanding of current coding guidance are taking on real compliance risk. The good news is that the structure for coding AI services is beginning to take shape and radiology professionals who understand that structure now will be far better positioned than those who wait.
What AI in Radiology Actually Means for Billing
Before we can talk about coding, it is worth pausing to clarify what we mean by AI in a billing context. In the broadest sense, AI in diagnostic radiology refers to software that analyzes imaging data, whether from CT, MRI, mammography, X-ray, or other modalities, to provide clinical decision support, quantitative measurements, or detection assistance. Breast imaging has served as a testing ground across generations of AI, from early computer-aided detection (CAD) systems to today's deep learning models. For new technologies, the key billing question is whether that software analysis constitutes a separately reportable service or whether it is considered part and parcel of the underlying imaging study.
The answer, increasingly, is it depends on the code.
CMS and the American Medical Association’s (AMA) CPT Editorial Panel have taken a case-by-case approach to AI services, evaluating each technology on its own merits before allowing the creation of procedure codes. Some AI applications have earned Category I CPT codes, reflecting widespread adoption and established clinical utility. Others remain in Category III, the CPT tracking category used to collect data on emerging services. Many AI applications currently have no billable code at all, meaning they cannot be reported as a separate service regardless of the clinical value they provide.
Category I: Where AI Has Already Landed
The most clinically significant AI coding development in recent years involves quantitative coronary atherosclerotic plaque analysis. This web-based service analyzes coronary computed tomographic angiography (CTA) datasets using algorithms to assess the extent and severity of coronary artery disease, with findings provided to the physician who performs the final review and interpretation.
For 2026, CMS deleted the Category III codes that had tracked this service (0623T through 0626T) and replaced them with a single new Category I code:
75577 — Quantification and characterization of coronary atherosclerotic plaque to assess severity of coronary disease, derived from augmentative software analysis of the data set from a coronary computed tomographic angiography, with interpretation and report by a physician or other qualified health care professional
The transition from Category III to Category I is meaningful for several reasons. It signals that this technology has crossed the threshold of being considered experimental or investigational from a coding standpoint. It also means that the service is now subject to standard billing rules and standard audit scrutiny.
A few critical operational points deserve emphasis. Code 75577 is reported once per coronary CTA, regardless of the number of algorithms applied or the complexity of the findings. It does not include the interpretation and report of the coronary CTA itself, which is separately reported with code 75574. When both services are performed on the same date of service, they may be reported together but organizations must ensure that documentation supports both the coronary CTA interpretation and the plaque analysis report as distinct physician work products.
Similarly, computed fractional flow reserve (FFR) derived from coronary CTA data is separately reportable under code 75580. This service involves uploading the dataset to a vendor, having the analysis performed, and receiving a report back to the physician. Like 75577, it is reported once per coronary CTA. Organizations offering both plaque analysis and FFR should review their workflows carefully to ensure that each service is supported by its own documentation and that billing for these distinct analyses is accurate.
Category III: Tracking the Future
While Category I procedure codes represent services that have widespread adoption, Category III codes allow for tracking of newer technologies that are still building their evidence base. In radiology, several AI-related Category III codes are currently in use and organizations performing these services should be reporting them, both to capture whatever revenue is available and to contribute to the utilization data that will inform future Category I status determinations.
A sampling of Category III AI codes of interest in radiology include:
Perivascular fat analysis (0992T and 0993T): These codes describe the use of AI software to analyze perivascular adipose tissue from existing CT images, providing quantitative measurements that may serve as markers of cardiovascular risk. The fact that these codes use previously acquired CT data raises an important workflow point. The CT itself must have already been performed, and the images must be available for the AI analysis to occur. There is no separate technical component (facility charge) being billed for the imaging when these codes are reported.
Chest imaging AI (0877T through 0880T): This family of codes covers AI-based analysis of chest radiographs and CT images for detection of various abnormalities. Organizations using AI software to flag or prioritize chest imaging findings should determine whether the specific software they are using maps to any of these codes and whether they meet the documentation standards required to report them.
Prostate biopsy AI (0898T): This code describes the use of AI software in conjunction with MRI-guided prostate biopsy procedures. As prostate biopsy coding has undergone significant restructuring with the new 2026 codes, organizations should ensure that the use of AI assistance in target identification is being captured appropriately when applicable.
On the Horizon: AI That Cannot Yet Be Billed
Some of the most clinically valuable AI applications in radiology currently have no CPT code at all. Pulmonary embolism detection AI, new mammography AI technologies, and intracranial hemorrhage detection tools are increasingly deployed in real-world practice flagging critical findings, triaging worklists, and expediting care but they do not yet have an established reporting mechanism under Medicare or most commercial payers.
This creates an awkward operational reality. Organizations are investing in these tools and integrating them into clinical workflows, but they cannot currently bill for the AI component separately. In most cases, the cost of the software is considered part of the overhead of providing the underlying imaging service. That may change as CMS and the CPT Editorial Panel continue to evaluate these technologies. Radiology professionals should be paying attention to Category III code additions in the upcoming months and years as that is typically where new AI services make their first appearance.
Documentation: The Non-Negotiable Foundation
Regardless of which AI code is being reported, documentation remains the non-negotiable foundation of compliant billing. This is an area where AI in radiology introduces new considerations that do not have a well-established precedent.
For most diagnostic imaging services, the radiologist's dictated report is the primary support for the procedure code that is billed. The same principle applies to AI services, but what the report needs to contain is less intuitive. For a service like coronary plaque analysis (75577), the physician must produce an interpretation and report that is separate and distinct from the coronary CTA interpretation. Simply referencing the AI output within the primary CTA report is not sufficient if the intent is to bill for both services. The plaque analysis report must stand on its own as a discrete physician work product.
This principle extends to other Category III AI codes as well. If your organization is reporting chest imaging AI codes, the documentation must reflect that the AI analysis was performed, what it found, and most critically that a physician reviewed the output. The physician's involvement in reviewing and interpreting AI-generated findings is a prerequisite for reporting these services, and that involvement must be evident in the record.
As a practical matter, many organizations are finding it useful to develop documentation templates that guide radiologists through the elements needed to support AI code billing. The template does not write the report. The physician's clinical judgment and specific findings do that, but it ensures that the structural elements required for billing are consistently present.
Compliance Monitoring in an Emerging Space
One of the challenges of coding for AI services is that audit guidance is still developing. National Correct Coding Initiative (NCCI) and other payer edits, payer policies, and CMS coverage determinations for AI services are evolving alongside the technology itself. That means organizations cannot simply apply historical audit frameworks and assume they are covered.
A few monitoring priorities stand out. First, organizations should track denial rates specifically for AI codes, separated from the underlying imaging services. If denials are clustering around a particular code or payer, that is a signal worth investigating. It may indicate a documentation gap, a coverage policy the organization was not aware of, or a coding error in how the AI service is being linked to the primary study.
Second, organizations should periodically verify that the AI software they are using actually maps to the CPT codes they are reporting. Software vendors sometimes market their products using billing codes without fully accounting for the specificity of the procedure descriptor. The organization, not the vendor, is responsible for accurate code assignments, and that responsibility requires understanding exactly what the software is doing and whether it aligns with what the code describes.
Third, any time a new AI service is introduced or a new Category III or Category I code becomes effective, organizations should treat it with the same operational rigor it would apply to any major coding change: update the charge description master (CDM), modify order sets and radiology information system (RIS) workflows, provide targeted education to coders and technologists, and implement early monitoring to catch issues before they compound.
Looking Ahead
The pace of AI development in radiology is not slowing down. New applications are being introduced faster than CPT codes can be assigned, and CMS is being asked to make coverage and payment decisions about technologies that were not on the market when the current reimbursement frameworks were built. That is not a comfortable regulatory space, but it is where radiology professionals are operating in right now.
What has not changed is the underlying principle that drives everything else: Procedure codes translate directly into dollars that affect our patients, our providers, and the financial health of our organizations. Accurate coding for AI services is not just a compliance issue, it is a revenue integrity issue and, increasingly, a strategic one. Organizations that build the operational infrastructure now to code AI services correctly will be better positioned to capture reimbursement as these technologies continue to mature and additional codes come online.
Stay curious, stay current, and make sure your stakeholders are along for the ride. The door to AI reimbursement in radiology is opening and it pays to know what is on the other side.
DISCLAIMER: CPT® is a registered trademark of the American Medical Association. CPT® five-digit codes, nomenclature, and other data are copyright 2025 American Medical Association. All Rights Reserved. No fee schedules, basic units, relative values, or related listings are included in the CPT® book. AMA does not directly or indirectly practice medicine or dispense medical services. AMA assumes no liability for the data contained herein or not contained herein.