Radiologists reading a 3D mammogram have always been expected to compare the current scan against a patient’s prior studies, looking for subtle changes over time that can signal an emerging cancer. Doing that comparison by hand, flipping between years of tomosynthesis images, is slow and easy to do inconsistently. On March 26, 2026, the FDA granted 510(k) clearance to version 1.2 of Lunit INSIGHT DBT, a South Korean company’s AI algorithm for digital breast tomosynthesis, adding a feature that automates exactly that comparison.
What’s New in the Updated Clearance
The marquee addition in version 1.2 is current-prior comparison, which allows the software to pull up AI findings not just for the mammogram being read that day, but for up to two prior exams as well, surfacing the comparison directly within the radiologist’s workflow rather than requiring a separate manual review step. The update also introduces multiple selectable operating thresholds, letting individual hospitals and imaging centers tune the balance between sensitivity and specificity to match their own patient population and callback tolerance, and it folds in volumetric breast density data alongside the AI’s cancer-detection findings, a feature relevant given the growing number of states that require density notification to patients.
In its FDA submission, Lunit argued that version 1.2 is substantially equivalent to the predicate device already on the market, since the new features don’t change the system’s intended use or introduce new questions of safety or effectiveness. Performance testing submitted alongside the application showed the updated version performs as safely and effectively as its predecessor at detecting suspicious lesions on tomosynthesis exams from compatible scanners.
A Platform Already Running at Scale
Lunit INSIGHT DBT isn’t a new entrant to the U.S. market; the underlying product first won FDA clearance in 2023, and the company has used the years since to expand its footprint substantially. By early 2026, Lunit said its broader breast-imaging AI suite had surpassed 330 sites and was processing more than one million mammography screenings annually, according to the company’s own reporting on its clinical adoption. The United States represents the largest single market opportunity for 3D breast imaging AI, accounting for more than 64% of global demand for the technology by Lunit’s estimate, which explains why incremental U.S. clearances like this one carry outsized commercial weight for the company even though the underlying technology isn’t brand new.
Why Current-Prior Comparison Matters Clinically
Breast cancer screening relies heavily on interval change, the difference between how a region of tissue looked last year and how it looks now, as one of the most reliable signals of an emerging malignancy. Radiologists already do this comparison manually as standard practice, but manual comparison is time-consuming, especially in high-volume screening practices reading dozens of studies a day, and it is vulnerable to the kind of fatigue-driven inconsistency that has been documented in radiology reading studies for decades. By automating the retrieval and overlay of prior AI findings, Lunit’s update is aimed squarely at that bottleneck, potentially shaving meaningful time off each read without asking radiologists to change how they already practice.
The Adjunctive Caveat
Like essentially every FDA-cleared mammography AI tool on the market, Lunit INSIGHT DBT is cleared as an adjunctive tool, meaning it is meant to support a radiologist’s interpretation rather than issue an independent diagnosis. That distinction matters for liability and workflow design: hospitals deploying the tool still need a qualified radiologist reviewing every study, and the AI’s abnormality scores and highlighted lesions are explicitly framed as a second check rather than a replacement read. Some radiologists have welcomed configurable operating thresholds as a meaningful improvement, since a one-size-fits-all sensitivity setting can produce very different callback rates depending on a population’s underlying cancer prevalence and the density mix of patients at a given site. Others in the field have cautioned that giving individual sites control over sensitivity thresholds introduces variability that could make it harder to compare AI-assisted screening outcomes across institutions.
What’s Next for AI-Assisted Mammography
Lunit’s update lands amid a broader expansion of FDA-cleared breast imaging AI, including competing current-prior comparison features from other manufacturers and separate clearances for tools that estimate individual breast cancer risk from mammogram images rather than simply flagging suspicious lesions. As more imaging centers adopt AI-assisted screening as a baseline expectation rather than a premium add-on, incremental software updates like version 1.2 are likely to become a regular cadence rather than a rare event, with companies competing on workflow refinements as much as on raw detection accuracy. For hospitals already running Lunit’s platform, the update is expected to roll out as a software upgrade to existing installations rather than requiring new hardware.
Photo: rayvision / PIXABAY via Pixabay
