Imagine getting a clean bill of health after your annual mammogram, only to find out months later that you have breast cancer. This situation is more common than many realize, and it’s known as an “interval breast cancer” — a cancer that shows up between regular screenings, even though the last mammogram didn’t raise any red flags.
A recent retrospective study led by researchers at UCLA dug deep into this issue, analyzing how and why these cancers are missed — and whether artificial intelligence (AI) can help catch them earlier. The findings offer hope for improving breast cancer detection and reducing those nerve-wracking cases that fall through the cracks.
What Is an Interval Breast Cancer?
Interval breast cancers (IBCs) are diagnosed after a normal screening mammogram but before the next one is due. In the U.S., many women are screened annually, so these cancers typically show up within a 12-month window. These cancers can grow quickly and tend to be more aggressive than those caught during routine screenings, which makes catching them early all the more important.
But here’s the tricky part: sometimes the cancer was visible on the previous scan, but it was so subtle that it was missed. Other times, it truly wasn’t visible yet. Understanding these differences is crucial for improving detection.
The UCLA research team analyzed nearly 185,000 mammograms taken between 2010 and 2019. They identified 148 women who were diagnosed with breast cancer less than 12 months after a “normal” mammogram.
Then, a team of experienced breast radiologists reviewed those prior “normal” scans again — but this time with the knowledge that the patient was later diagnosed with cancer. They grouped each cancer into one of six categories based on whether anything suspicious was visible on the original scan:
- Missed–Reading Error: Clearly visible but overlooked by the radiologist.
- Minimal Signs – Actionable: Subtle signs that an expert might have flagged.
- Minimal Signs – Non-Actionable: Subtle signs no radiologist could reasonably catch.
- True Interval Cancer: Not visible on the previous scan — it developed between screenings.
- Occult: Still not visible even after diagnosis, but picked up with other imaging like ultrasound or MRI.
- Missed–Technical Error: The cancer wasn’t captured properly due to technical issues (like poor image quality or positioning).
These categories help paint a clearer picture of how interval cancers sneak by — and where there might be room for improvement.
Enter Artificial Intelligence
The big question this study tackled: Can AI help detect cancers that human eyes miss?
To test this, researchers used an FDA-cleared AI tool called Transpara to reanalyze the old mammograms that were previously considered normal. This AI assigns each image a score from 1 to 10, reflecting the likelihood of cancer being present. A score of 8 or higher is considered high risk and would normally trigger closer scrutiny.
Here’s what they found:
- AI flagged 76% of the interval cancers overall.
It was especially good at picking up on cancers that were actually visible in hindsight — scoring:
- 90% of missed-reading error cases,
- 89% of minimal signs-actionable,
- 72% of minimal signs-non-actionable.
But it was less effective at spotting:
- 50% of true interval cancers,
- 69% of occult cancers,
- and only 40% of missed-technical errors.
Translation? The AI was pretty good at catching things that could have been seen the first time — even if they were subtle — but not so great at detecting cancers that genuinely weren’t there yet.
One of the most encouraging takeaways is that almost two-thirds of the interval cancers in the study were “mammographically visible,” meaning there were subtle signs present that could be spotted — either by a very experienced radiologist or, perhaps more consistently, by AI.
In fact, if AI had been used alongside radiologists during the original screenings, the researchers estimate that up to 30% of these missed cancers might have been caught earlier.
That could mean fewer aggressive cancers going undetected, less invasive treatment, and better outcomes for patients.
Not All Cancers Are Equal
The study also found some patterns worth noting:
- Younger women (age 40-49) were more likely to have occult cancers — those that didn’t show up on mammograms but were later detected with ultrasound or MRI. These women also tended to have denser breast tissue, which can make mammograms harder to interpret.
- Triple negative breast cancers — an aggressive type that doesn’t respond to hormone therapy — were more likely to fall into the “true interval” group. That means they probably weren’t visible yet on the mammogram and developed rapidly.
- Hormone receptor-positive cancers were more likely to be the subtle, slow-growing types that might be flagged with AI help.
So, Is AI the Solution?
It’s tempting to think AI could be the silver bullet — but it’s not quite that simple.
AI tools are trained on thousands of images and can help catch things human eyes might miss. But they also tend to flag about 30% of all mammograms for follow-up, which would be too many if doctors treated them all as urgent.
In this study, the AI marked the correct location of the cancer less than half the time when it did flag an exam. And while it excelled at catching certain types of visible cancers, it still struggled with ones that truly weren’t there yet.
So, while AI is promising — especially as a second pair of eyes — it’s not ready to replace human judgment. It’s a support tool, not a crystal ball.
The researchers stress that this was a retrospective study — they were looking at old scans and checking to see how AI would have done. What we don’t know yet is how well it performs in real-time practice or how radiologists will actually use its findings in the moment.
We also don’t know how many of the AI-flagged cases would have led to follow-up imaging or biopsies — and whether that would’ve resulted in earlier diagnoses without increasing false alarms.
Still, this research marks an important step forward. It shows that a significant number of interval cancers — the ones that sneak in between screenings — are visible in hindsight, and that AI might be a helpful ally in spotting them earlier.
If you’ve ever worried about a mammogram missing something, you’re not alone — and researchers are working hard to make screening more accurate and effective.
Here are some takeaways:
- Don’t skip your mammogram. Annual screenings are still the best defense we have for early detection.
- Ask about breast density. If you have dense breasts, talk to your doctor about whether additional imaging like ultrasound or MRI might be helpful.
- AI is on the rise. While it’s not yet standard everywhere, tools like the one in this study are already being tested and implemented in some screening centers.
- Early detection matters. The goal is to catch cancers when they’re small and more treatable — and studies like this are helping move the needle in the right direction.
Breast cancer screening isn’t perfect, but it’s improving — and artificial intelligence may play a key role in that progress. By identifying the small signs that humans might miss, AI has the potential to help radiologists catch more cancers earlier, especially the ones hiding in plain sight.
