Radiology Coding Alert

Remember to review records regardless of what created them.

Artificial intelligence (AI) is being used throughout healthcare — from detecting abnormalities on radiology images to listening to visits to produce notes for the electronic health record (EHR) and suggesting codes to report based on the documentation. During her “Smart Compliance: Leveraging AI Without Compromising Accountability in Medical Coding & Auditing” session at AAPC HEALTHCON 2026, Ivonne Atanacio, CPC, CPB, CPMA, CRC, CPC-I, CEMC, CGSC, CASCC, CEO/founder of A16 Coding, broke down where AI is helpful in the revenue cycle and examined the risks of the technology.

Read on to find out what Atanacio had to say about AI’s role in medical billing and why your insight is as important as ever.

Understand Where AI Is Being Deployed

Some practices are hesitant to deploy the technology in their organizations, but AI is here, and recognizing it as an assistive tool will benefit your business.

“I started thinking, it’s almost like swimming against the waves. You’re not going to get far if you swim against it, but if you swim with it, things will get better. So, I had to reframe my thought process when it came to AI,” Atanacio explained.

Regardless of where you stand on the technology, if and when you’re ready to acclimate yourself to AI, it’s important to educate yourself on the tool. You’ll need to understand what you’re working with, so you can take advantage of its capabilities.

“You’re going to want to understand how AI truly helps you and your organization; you might feel a little bit more confident when you’re working with AI, but be cautious by using it and always, always, always, making sure that you’re protecting your organization, yourself, and your patients. That is our job,” Atanacio warned.

In radiology, AI is currently used to compare mammogram images and identify brain injuries on MRI exams, but AI’s role has also expanded into coding, billing, and auditing workflows over the years. The technology can provide coding assistance or be used as a coding suggestion tool, which are commonly known as computer-assisted coding (CAC).

Medical coders and billers might use AI to scrub their claims and for error detection before the claim heads to the payer for review. Professionals also use the technology to improve and generate documentation.

In each of these areas, AI can boost efficiency and automation across the organization. Tasks that may have consumed an entire workday or multiple workdays are now completed in a fraction of the time. On the surface, this does make your work life somewhat easier, but human oversight is still necessary. You are responsible for your work on a claim when it leaves the door bearing your seal of approval.

Making sure the AI’s code suggestions are accurate is critical because your payers are also using the technology. “If you work in the billing side, you are fighting a war because payers are just as equally, if not more, advanced than your organization,” Atanacio said. She argued that the payers’ use of automation is why you’re experiencing a higher number of denials and seeing more medical documentation requests.

Recognize the Risks of AI

As helpful as AI can be, the technology is not without its faults. Using AI in healthcare comes with coding, billing, auditing, documentation, and overreliance risks.

Coding risks: The issue with relying solely on AI-generated code suggestions is that the recommended codes might not reflect the true complexity of the patient’s conditions. Ambient AI can listen in on a visit while the physician speaks with the patient, but it may put extraneous information in the record that isn’t necessary; or, an AI coding software may suggest codes for conditions that happened in the past or are “probable.”

“You work very hard as a coder. You have the skill set, you’re applying it. Don’t over-rely on what automation is giving you. Fact check it. Fact check each other,” Atanacio said.

Billing and audit risks: AI technologies are not one-size-fits-all — they need to be customized and personalized to meet the needs of the organization that is using them. An AI system designed to identify abnormalities in mammograms might not fare as well when used to identify live images on an ultrasound.

The same reasoning applies to AI coding software, as the suggested codes or guidelines may not line up with payer-specific requirements. For example, a payer may require laterality modifiers like LT (Left side), RT (Right side), or 50 (Bilateral procedure) for imaging procedure codes, but if your AI technology does not have that information, it will miss it and you could receive a denial.

Documentation risks: Generative AI models like ChatGPT and Claude can produce authentic-looking medical reports in various formats, but a closer inspection can reveal that the notes are inaccurate. AI-generated documentation can also look complete, but it can lack clinical specificity.

Example: A patient with a history of breast cancer (in remission) presented to the radiologist for a diagnostic mammogram. The AI-generated documentation misinterpreted the reason for the visit and noted the patient was being seen for imaging of a current breast cancer condition. The coder reported an applicable C50.- (Malignant neoplasm of breast) code instead of Z85.3 (Personal history of malignant neoplasm of breast).

The inaccurate documentation will affect the reimbursement for the provider and medical care for the patient; it also puts the practice at risk of an audit and compliance violation.

Overreliance risks: Overreliance on AI may reduce critical review and validation. “From a personal standpoint, overutilization takes away from your critical thinking skills,” Atanacio said.

If you accept what the technology provides at face value and you don’t validate the codes suggested, that choice will impact accuracy going forward. “The problem with not monitoring the output of a large language model [LLM] is that you tend to trust it so much and then when you find one error because no one was looking at it, now you have 50 errors coming out,” Atanacio explained.

Simple errors, like a laterality mistake, can be duplicated and disseminated across multiple workflows.

Remember the Importance of Human Oversight

AI doesn’t create risk, but lack of oversight does. “At the end of the day, AI is not the problem. The problem is when we start reviewing it, validating it, and taking ownership of it. And in order to do that, it is our responsibility to proactively do something about it,” Atanacio said.

Regardless of whether your practice does or does not use AI in medical coding and billing, human oversight is essential. The technology should not replace professional judgment. It is your responsibility to know the payer policies, coding guidelines, and edits — and to review documentation during the coding process.

Without oversight over the automation, you’re going to have more problems than solutions.

Mike Shaughnessy, BA, CPC, Production Editor, AAPC