The Lunit INSIGHT DBT's algorithm analyzes 3D mammograms slice by slice, highlighting potential abnormalities. (Photo: Lunit)
Despite decades of progress in breast imaging, one challenge continues to test even the most skilled radiologists: interval cancers — breast cancers that appear between regular screening rounds after an initial “negative” mammogram. Often fast-growing and aggressive, these cancers can account for a significant proportion of clinically relevant misses, even in the era of digital breast tomosynthesis (DBT).
While DBT has advanced detection rates and reduced recalls, it hasn’t eliminated every blind spot. The question remains: can technology help radiologists see what the human eye sometimes cannot?
A new study published in Radiology1 and led by Dr. Manisha Bahl, Associate Professor of Radiology at Harvard Medical School and Associate Medical Director of Quality at Mass General Brigham, offers compelling evidence that it can.
In an interview with The Imaging Wire, Dr. Bahl discussed her team’s findings using artificial intelligence (AI) — specifically Lunit INSIGHT DBT®: “Our study shows that an AI algorithm can retrospectively detect and correctly localize nearly one-third of interval breast cancers on screening DBT exams, suggesting its potential to reduce the interval cancer rate and improve screening outcomes.”
The findings highlight a powerful new role for AI in breast imaging — not as a replacement for radiologists, but as an additional lens for earlier, more confident detection.
The Persistent Challenge of Interval Cancers
Interval cancers are diagnosed after a “negative” screening exam and before the next routine screening — often within 12 months. They are clinically important for two reasons: they tend to be more aggressive, and they signal potential limitations in current imaging performance.
Dr. Bahl explained, “The interval cancer rate is an important metric — it’s thought to be a surrogate marker for long-term outcomes, especially since we lack long-term data on the impact of tomosynthesis on mortality.”
For Dr. Bahl, this challenge raised an important research question: could AI help radiologists identify these cancers earlier — before they become symptomatic, and before the window of opportunity closes?
The Study: Putting AI to the Test
To explore this question, Dr. Bahl’s team conducted a retrospective analysis of more than 200 interval cancer cases for which the initial screening DBT exams had been interpreted as negative. Each screening mammogram was analyzed using Lunit INSIGHT DBT, a deep learning–based algorithm trained on millions of annotated mammographic images from diverse populations.
Unlike traditional computer-aided detection (CAD) tools, which rely on human-engineered rules, Lunit INSIGHT DBT learns directly from image data — identifying subtle patterns and relationships that may not be perceptible to human readers.
In the interview, Dr. Bahl emphasized the study’s approach: “We wanted to know whether AI could detect and correctly localize the site of cancer, not just assign a high exam score.”
The results were impressive.
- The AI detected and correctly localized nearly one-third of all interval cancers at the time of screening.
- Interval invasive cancers detected by AI were significantly larger at the time of surgery and more frequently lymph node positive.
- These results suggest that AI preferentially detects more clinically advanced tumors that, if caught earlier, may have been smaller and of a lower stage.
Two cases illustrate the real-world impact of these findings:
- In one case, a 41-year-old woman’s screening digital breast tomosynthesis exam was read as negative. Ten months later, the patient presented with a lump in the left breast and was subsequently diagnosed with grade 3 invasive ductal carcinoma. Upon retrospective evaluation of the initial screening mammogram, the AI algorithm correctly localized the cancerous distortion with a high score.
- In another case, a 48-year-old woman presented for a screening digital breast tomosynthesis exam; its findings were interpreted as negative. Eleven months later, the patient presented with a lump in the left breast and was subsequently diagnosed with grade 1 invasive ductal carcinoma. Upon retrospective evaluation of the initial screening mammogram, the AI algorithm marked the mass with associated architectural distortion correctly with a high score.
The authors conducted an additional analysis of 1,000 screening exams to further contextualize the broader performance of the AI algorithm. The AI accurately localized 84.4% of screening-detected cancers and correctly categorized the majority of true-negative (85.9%) and false-positive (73.3%) cases as normal.
Dr. Bahl reflected on the implications: “We’re increasingly integrating AI-based detection and diagnosis tools into clinical practice. Retrospective studies like ours are promising, but we’ll need to monitor real-world performance metrics — such as cancer detection and interval cancer rates — after AI deployment to truly understand its impact.”
Why Lesion-Level Validation Matters
In breast imaging, location accuracy isn’t just a detail — it’s the difference between theoretical and clinical value. Many earlier AI studies relied on exam-level metrics: if AI flagged anything and cancer existed somewhere in the breast, it counted as correct. But that’s not how radiologists work.
Dr. Bahl highlighted this level of precision in the Imaging Wire interview: “We’re among the first to rigorously confirm lesion-specific accuracy with DBT, which provides more meaningful evidence of AI’s clinical value.”
This precision gives radiologists confidence that AI is marking clinically relevant areas. In an accompanying Radiology commentary2, Dr. Christopher Lee and Dr. Hannah Milch noted that lesion-level validation “provides more meaningful insight into AI’s potential to impact real-world screening.”
Beyond Detection: The Broader Role of AI in DBT
AI’s value in breast imaging extends well beyond early diagnosis and minimizing interval cancer rates. It can also make DBT — already a rich but time-consuming modality — more efficient and consistent.
Dr. Bahl told The Imaging Wire, “Digital breast tomosynthesis has become the standard of care in the U.S., offered at over 90% of facilities. But with more image slices to review, interpretation time increases. AI can help improve efficiency by analyzing large volumes of data quickly and flagging suspicious findings, supporting radiologists in high-volume screening settings.”
While AI currently serves an assistive role in breast imaging, its capabilities continue to advance. Increasingly, both clinicians and patients recognize AI’s value as a partner in care — acting as a second reader that enhances accuracy, consistency, and confidence in screening outcomes.
From Research to Real-World Impact
Dr. Bahl’s team also is exploring how AI can guide treatment decisions, particularly for women with stage 0 or high-risk lesions. “In active surveillance trials for stage 0 breast cancer — where patients are monitored instead of undergoing surgery or radiation — AI could help risk-stratify which women can safely avoid aggressive treatment,” she said.
This vision reflects a broader shift in oncology: using AI not only to find cancer earlier, but to personalize care across the continuum — from screening to management.
The Role of Lunit INSIGHT DBT
At the center of this research is Lunit INSIGHT DBT, the AI tool used in Dr. Bahl’s study. The algorithm analyzes 3D mammograms slice by slice, highlighting potential abnormalities such as architectural distortions, masses, asymmetry, and calcifications.
Dr. Bahl explained to The Imaging Wire, “We analyzed over 200 interval cancer cases — all digital breast tomosynthesis exams initially interpreted as negative. We then processed them using the Lunit INSIGHT DBT algorithm to see if AI could have detected these cancers earlier.”
Developed using millions of global, annotated images, the software is designed for high sensitivity and precise lesion localization — attributes that made it ideal for this research. Already deployed in clinical programs worldwide, Lunit INSIGHT DBT supports radiologists in high-volume environments, helping reduce false negatives and boost confidence in subtle cases.
Seeing What Others Can’t
Interval cancers remain one of the most challenging aspects of breast screening — elusive, often aggressive, and difficult to predict. But as this study shows, AI is offering a new lens through which radiologists can detect what was once unseen.
By combining human expertise with machine precision, tools like Lunit INSIGHT DBT are helping redefine what’s possible in breast imaging. The evidence is growing: AI isn’t just about efficiency — it’s about meaningful impact on patient outcomes.
See Lunit INSIGHT DBT in action at RSNA 2025. Book a demo now.
References
- Bahl M et al. AI to Reduce the Interval Cancer Rate of Screening Digital Breast Tomosynthesis. Radiology. 2025 Jul;316(1):e241050
- Lee C et al. Clinically Meaningful AI Detection of Interval Breast Cancer at Digital Breast Tomosynthesis Screening. Radiology 2025; 316(1):e251860
Reference — [For U.S.] Indications for use: Lunit INSIGHT DBT is a computer-assisted detection and diagnosis (CADe/x) software intended to be used concurrently by interpreting physicians to aid in the detection and characterization of suspected lesions for breast cancer in digital breast tomosynthesis (DBT) exams from compatible DBT systems. Through the analysis, the regions of soft tissue lesions and calcifications are marked with an abnormality score indicating the likelihood of the presence of malignancy for each lesion. Lunit INSIGHT DBT uses screening mammograms of the female population.
Lunit INSIGHT DBT is not intended as a replacement for a complete interpreting physician’s review or their clinical judgment that takes into account other relevant information from the image or patient history.
Disclaimer:
The product described is CE-marked and complies with the EU Medical Device Regulation (MDR) 2017/745 and applicable EU standards. It is also FDA-cleared.
Lunit INSIGHT DBT is a Computer-Assisted Detection/Diagnosis (CADe/x) software intended to be used concurrently by interpreting physicians to aid in the detection and characterization of suspicious lesions for breast cancer in digital breast tomosynthesis (DBT) exams. The device is an adjunctive tool, supporting readings for interpreting physicians.
This document is for use by healthcare professionals only. The radiologist should always rely on his or her own clinical and professional opinion when deciding whether to use a certain product to diagnose or treat a patient.
Availability of Lunit products may vary by market, depending on local medical and/or regulatory requirements. Please contact your Lunit representative if you have questions about the availability of the Lunit products in your area.
Lunit Inc. owns, uses or enforces the following trademarks or service marks: Lunit, Lunit INSIGHT.
Lunit Inc., 4-9F, 374, Gangnam-daero, Gangnam-gu, Seoul, 06241, Republic of Korea
October 21, 2025 
