Making Dollars and Sense of AI in Breast Imaging: The Economic Reality Check

After spending the last six months helping practices implement AI solutions, I've noticed something: everyone wants to talk algorithms and detection rates, but hardly anyone's tackling the elephant in the room—who's actually paying for all this?

Having started my career implementing CAD and EHR systems in hospitals and imaging clinics back in the "meaningful use" days of 2011, I've seen this movie before. The technology changes, but the economic challenges remain eerily similar. Let me share what I've learned from over a decade in the trenches.3

The Ghost of CAD: Why History Matters for Your Wallet

Remember CAD? When it first rolled out in the late '90s, practices could bill separately for it—a nice little revenue stream for early adopters. I still remember a practice in Colorado showing me their spreadsheet of how CAD was actually making them money back then.

Fast forward to today? That CAD payment got bundled into the mammography code itself. You now get "paid" for CAD whether you use it or not, but here's the kicker—that bundled amount doesn't come close to covering modern systems' costs.

This isn't just ancient history—it's exactly what could happen with today's AI tools. One radiologist I work with called this the "innovation penalty" in healthcare: adopt early, pay more, eventually get bundled, repeat cycle.

Medicare's Split Personality on AI Reimbursement

Here's something I've had to explain to every practice thinking about AI investments: not all AI tools have the same reimbursement potential. It breaks down like this:

Detection AI: Tools that highlight findings radiologists could potentially see themselves (basically CAD on steroids). I've implemented several mammography AI detection tools that are genuinely impressive, but good luck getting paid specifically for using them.

Quantification AI: These provide information radiologists can't derive independently—like those ultrasound tools that spit out malignancy risk scores. A radiologist in Texas told me, "I can't calculate those probability scores in my head—that's net new information."

This distinction is EVERYTHING when it comes to getting paid. Medicare has shown they're willing to fork over money for quantification tools because they deliver novel information. Some of these tools in the hospital outpatient setting are reimbursed at rates higher than the imaging exam itself!

Meanwhile, detection algorithms remain mostly unreimbursed, despite vendors swearing up and down that reimbursement is "just around the corner." Trust me, I've sat through those sales pitches too.

Behind the Curtain: How AI Companies Are Evolving Their Business Models

I've watched AI vendors get smarter about their business models. Most started with subscription pricing (one annual fee, unlimited use). Now? Almost everyone's switched to per-case fees.

This isn't random—it's strategic positioning within Medicare's reimbursement structure. Here's what a vendor CEO of an imaging explained to me recently:

Medicare distinguishes between "indirect" practice expenses (your rent, staff, equipment) and "direct" expenses tied to specific patients. The per-click model (per-case pricing) aligns with direct expenses, which typically receive better reimbursement. Smart move on their part, frustrating reality for budget-constrained practices.

The Real Work: What Radiologists Do With AI (That Nobody's Paying For)

Nobody seems to be talking about this enough: using AI creates actual work for radiologists. It's not just "push button, get answer."

When a radiologist uses AI, they need to:

· Review what the algorithm found

· Decide whether they agree with it

· Figure out how to incorporate it into their report

· Stand behind that decision medicolegally

One Women’s Breast radiologist I work with tracked her time and found that critically evaluating AI findings added about 45 seconds per case. Multiply that by a busy practice's volume, and you're looking at significant uncompensated work.

The ACR is exploring an "algorithm-agnostic" code that would pay for this cognitive work regardless of which AI tool you're using. But until that happens? It's all on your radiologists' dime.

The Integration Problem Nobody Warned Me About

Here's something I learned the hard way helping a 5-radiologist practice implement their first AI tool: integration is BRUTAL for smaller groups.

Imagine trying to separately integrate dozens of individual AI algorithms into your workflow. One practice I consulted for tried implementing three different algorithms simultaneously—it was a workflow nightmare that almost tanked their productivity for a month.

The market is responding with platform solutions offering bundles of algorithms through a single integration point. For many practices, especially those without dedicated IT teams, this approach is the only viable path forward.

Hidden Costs That Blew Up My Client's Budget

When budgeting for AI, I always tell clients the sticker price is just the beginning. Here are the "surprises" I've seen derail ROI calculations:

Integration costs: One practice spent $22K just getting their new AI tool to play nice with their existing PACS. Their vendor conveniently left that part out of the sales pitch.

Algorithm oversight: Who's checking whether the AI is still performing correctly after six months? A client in the West Coast assigned this to their medical director, who quickly realized it was practically a part-time job.

Selection costs: The hours spent evaluating options is real time and money. One group had three radiologists spend five hours each reviewing different vendors—that's 15 hours of radiologist time nobody accounted for.

Training and workflow adaptation: Every radiologist adapts differently. I've seen this range from "picked it up in an hour" to "fighting it six months later."

Maintenance and updates: Just like any software, these tools need updating and occasional troubleshooting.

These "hidden" costs have made or broken the economic viability of several implementations I've overseen. The groups that plan for them succeed; those that don't end up with buyer's remorse.

The Time Paradox I Never Saw Coming

Here's my favorite AI irony: tools marketed as time-savers don't always deliver in real-world settings.

I watched a breast radiologist use a new AI tool that flagged a subtle finding. She spent nearly 8 minutes deciding whether she agreed with the algorithm's assessment—way longer than she would have spent on that case otherwise.

As she told me afterward, "If I disagree with AI, I better be damn sure about it, because if it turns out to be cancer, the first question in court will be 'why did you ignore the AI?'"

Add to this the workflow inertia—stopping to learn a new tool means falling further behind on the day's worklist—and adoption becomes far more complicated than vendors admit.

Value-Based Care: The Backdoor to Making AI Pay Off

Here's where things get interesting. In traditional fee-for-service, getting paid for AI use is an uphill battle due to budget neutrality constraints. But in value-based payment models? The equation changes entirely.

I've seen this firsthand with a group in Massachusetts operating under an alternative payment model where they're rewarded for keeping patients healthy rather than just doing more procedures-value-based care.

 For them, an AI tool that finds cancers earlier or reduces unnecessary follow-ups creates immediate financial value—whether there's a specific billing code or not.

This creates a fascinating dynamic where AI adoption is accelerating in regions where value-based care has more traction. I'm currently working with three different health systems in these regions who are investing heavily in AI despite minimal fee-for-service reimbursement.

Women's imaging centers and breast radiology practices are uniquely positioned to capitalize on this. As one of my clients put it: "Unlike other radiology subspecialties, we manage the entire screening pathway. We can actually measure and capture the benefits of finding cancer earlier.”

Finding Cancer Earlier: The Economic Argument That's Working

Here's a talking point that's resonating with hospital administrators: as treatments for advanced breast cancer become increasingly expensive (some running $150K+ per patient), finding cancer earlier through AI-enhanced screening creates substantial cost savings.

I recently helped a practice present this economic case to their hospital board. We calculated the potential savings from stage migration—finding more cancers at stage 1 rather than stage 3—and the numbers were compelling enough to secure funding for their AI implementation despite no direct reimbursement.

Where Do We Go From Here?

Based on everything I'm seeing in the field, here's where I think we're headed:

· Platform solutions will win over individual algorithms for all but the largest practices

· Reimbursement will eventually come for the cognitive work of using AI, but not for years

· Value-based care systems will adopt AI faster than fee-for-service holdouts

· The strongest economic case for AI will continue to be earlier detection and workflow efficiency

Your Turn: What's Happening in Your Practice?

I'm genuinely curious: How is your practice approaching the economics of AI implementation? Have you found creative ways to make the numbers work? Are you waiting for more established reimbursement pathways?

Drop me a line with your experiences. I'm compiling feedback for a more detailed analysis, and your perspective would be incredibly valuable.

Next month, I'll tackle "AI Performance Monitoring: Who's Watching the Algorithms?" based on what I'm seeing in practices nationwide. Stay tuned!