Radiology dominates FDA-cleared AI, but reimbursement lags far behind

The difference between CPT 1 and CPT 3 codes and payment

Rubin explained that the key issue is the difference between Category 1 CPT codes, which are used for payment, and Category 3 CPT codes, often referred to as “T codes,” which are temporary codes designed primarily to track the use of new or investigational technologies. When. AI vendors say they have a CPT code, most of the time they are referring to CPT 3 codes, which are not tied to payment.

“So, let me start with a little bit of the basics,” Rubin said. “CPT codes are what we use to describe the procedures that we perform and when we break them down.”

He said Category 1 CPT codes describe distinct procedures that are widely performed across the country and supported by a substantial body of clinical literature demonstrating their utility. These codes, once approved by the AMA CPT Editorial Panel, move on to the AMA Relative Value Update Committee (RUC), where physician work, time, intensity and practice expense are assessed to assign a relative value unit (RVU) used to calculate payments.

By contrast, Category 3 codes have a much lower bar for acceptance.

“They’re often investigational or representative or describing procedures that are relatively new or they’re not in widespread use,” Rubin said, adding that these codes “do not go to the RUC for valuation.”

According to Rubin, Category 3 codes exist so the AMA and specialty societies can understand how frequently a new technology is being used and whether it is gaining traction across multiple institutions. Only after sufficient utilization and literature support exist can a specialty society request conversion to a Category 1 code.

Only two CPT 1 codes exist for newer AI

That pathway has been successfully followed by only a few imaging AI applications. Rubin said that, at present, radiology has just one AI-related Category 1 CPT code for fractional flow reserve computed tomography (FFR-CT). A second for coronary plaque analysis on coronary CT angiography (CCTA) scans is currently a Category 3 code, but is scheduled to convert to Category 1 status on Jan. 1.

“Each of those procedures started as Category 3 CPT codes,” Rubin said. “But when they reached that bar, a request for conversion was submitted and accepted by the CPT editorial panel.”

But, he added that both FFR-CT and plaque analysis have a high level of clinical data showing improved outcomes and greater diagnostic value that CCTA exams on their own. And that clinical evidence is what helped propel these two technologies toward reimbursement.

Rubin said reimbursement potential can play a significant role in driving industry investment and adoption.

“When you start to see the potential opportunity for reimbursement, as you see with fractional flow reserve with plaque analysis, I think what you end up seeing is a greater degree of attention from industry, from the vendors in spending the money to bring forward those procedures in their individual companies for then use by the public and for selling those to the public,” he said.

This move toward reimbursement attracting additional vendor investment was evident on the RSNA show floor. At least two vendors were showing FFR-CT and at least three with plaque analysis AI. But there are numerous vendors developing their own versions of this technology, including two others at RSNA displaying FDA-pending versions of their plaque analysis software.

A sea of AI at RSNA with no reimbursement

However, Rubin cautioned that most AI algorithms on the RSNA expo floor are unlikely to ever receive their own paid CPT codes. These include tools for fracture detection, incidental findings or lung nodule identification. The reason, he said, is that radiologists are already reimbursed for interpreting imaging studies and are already supposed to be identifying those findings.

“What we are very, very careful about and what we understand in the CPT process is that CPT codes are meant to describe unique procedures that are performed,” Rubin said. Creating separate codes for tasks already inherent in existing imaging services would conflict with CPT rules and raise concerns about paying twice for the same work, he explained.

As a result, Rubin said the return on investment for many imaging AI tools may need to come from sources other than direct reimbursement, such as improved efficiency, reduced repeat imaging, or better downstream patient management at the health system level that helps reduce costs.

“The ROI, when you look at some of these technologies is not just to look at whether or not you get a one-to-one performance coding and payment for a procedure,” he said.

Despite the challenges, Rubin emphasized that engagement in the CPT process remains critical as imaging technology continues to evolve.

“A lot of work goes into the development of CPT codes,” he said, adding that sustained volunteer involvement is essential to ensuring new technologies can ultimately be integrated into a healthcare system with limited financial resources while still supporting high-quality patient care.