A groundbreaking study led by researchers at the UCLA Health Jonsson Comprehensive Cancer Center reveals promising advancements in breast cancer detection using artificial intelligence (AI). This research focuses on a particularly elusive subset of breast cancers known as interval cancers—tumors that develop and manifest in the time between routine mammographic screenings. By harnessing AI’s pattern recognition capabilities, this innovative approach aims to identify these cancers earlier, potentially transforming breast cancer screening protocols and improving patient outcomes in a significant way.
Interval breast cancers have historically posed a formidable challenge to radiologists. Unlike cancers detected during scheduled mammograms, interval cancers arise and are diagnosed after a negative screening and before the next recommended screening appointment. These tumors often grow aggressively, making early detection critical for effective treatment. What makes interval cancers particularly insidious is that they can either be missed during the initial mammogram due to faint or subtle indications or may not produce detectable signs at all, thereby escaping timely diagnosis.
The UCLA-led study, published in the Journal of the National Cancer Institute, analyzed nearly 185,000 mammograms collected over a decade, ranging from 2010 to 2019. This substantial dataset included images obtained from both digital mammography (DM) and digital breast tomosynthesis (DBT), the latter commonly known as 3D mammography, which is widely used in the United States. While most European screening programs rely on 2D digital mammography with intervals of two to three years, the U.S. approach tends to emphasize annual screenings and 3D imaging. Understanding AI’s applicability within this distinctly American clinical context adds critical value to this research.
At the core of their investigation was the application of Transpara, a commercially available AI software tool designed to evaluate mammograms and assign a cancer risk score ranging between 1 and 10. Scores of 8 or higher flagged a mammogram as potentially suspicious, prompting further radiological attention. The team retrospectively examined images from patients who were later diagnosed with interval cancers, using AI to reassess the mammograms initially read as normal to determine if subtle malignancy signals could have been detected earlier.
The findings are encouraging and demonstrate AI’s substantial potential to augment human diagnosis. The AI model flagged an impressive 76% of mammograms that were initially interpreted as cancer-free but were ultimately linked to interval cancers. This heightened detection rate suggests that AI could serve as a crucial second line of defense, identifying lesions that might evade even the most experienced radiologist’s eye. Particularly noteworthy is AI’s success in identifying "missed reading error" cases, where cancers were visible on the mammogram but overlooked, achieving a detection rate of 90%.
Moreover, AI performed admirably in detecting "minimal signs" cancers—tumors exhibiting subtle features that borderline on detectability. Approximately 89% of actionable minimal-signs cases were correctly flagged, meaning these are cancers showing slight but interpretable abnormalities that could reasonably prompt clinical intervention if noticed. The technology also showed promise in flagging non-actionable minimal-signs cancers, where signs were likely too inconspicuous to trigger immediate concern, correctly identifying 72% of such cases.
Even for occult cancers—tumors truly invisible on mammograms due to their nature—AI demonstrated an unexpected ability to flag 69% of those cases. This finding raises intriguing questions about whether machine learning algorithms can identify subtle imaging characteristics that transcend the visual limitations faced by human observers. However, this capability is tempered by AI’s relative struggle with “true interval cancers,” which genuinely develop in the interval between screenings and are not present during initial scans. AI flagged only about half (50%) of these genuinely new lesions, a reminder of the intrinsic difficulty in predicting tumors that rapidly emerge post-screening.
Despite these promising results, the study’s authors emphasize that AI is not a panacea and acknowledge significant limitations. For example, while the AI system flagged 69% of occult cancer mammograms, it managed to precisely pinpoint the actual cancer location only 22% of the time. This discrepancy between overall cancer suspicion and accurate lesion localization highlights a critical area for improvement before AI can reliably influence clinical decision-making at scale.
The research also underlines the necessity to investigate how integrating AI into routine screening workflows might influence radiologists’ interpretations and patient outcomes in real-world settings. There remain unresolved challenges, such as managing false positives and addressing cases where AI flags abnormalities that are imperceptible to human readers but may or may not represent clinically significant pathology. Determining appropriate responses to such AI alerts without causing unnecessary anxiety or interventions will require careful study.
“It’s a complex balance,” comments Dr. Tiffany Yu, assistant professor at UCLA’s David Geffen School of Medicine and the study’s lead author. “AI offers tremendous promise as a ‘second set of eyes,’ especially for the subtle, hard-to-detect cancers. But it still requires radiologists’ expertise to weigh these alerts and make the final call. Our findings suggest that incorporating AI could shift the profile of interval cancers more toward cases truly undetectable by imaging, which could ultimately save lives through earlier diagnosis.”
Senior author Dr. Hannah Milch further articulates the cautious optimism around AI’s role. While the technology exhibits impressive sensitivity for certain categories of interval cancers, it remains imperfect. The potential for AI to disrupt traditional screening methodologies is immense, but so too is the need for rigorous future research to refine AI algorithms, improve lesion localization, and map workflows that optimize collaborative human-machine decision-making.
This UCLA study stands among the first comprehensive explorations of AI’s role in interval breast cancer detection within the United States, addressing a clinical gap distinct from European populations where screening practices differ. These insights could drive tailored implementation strategies that harness AI’s strengths where they are most needed, ultimately enhancing screening efficacy in diverse healthcare settings.
Supported by funding from the National Institutes of Health, National Cancer Institute, and other agencies, this research signals a critical juncture in the ongoing evolution of breast cancer diagnostics. As AI systems become more sophisticated, they hold the potential to revolutionize the early detection landscape, offering hope for reducing breast cancer mortality by catching aggressive cancers before they escalate.
In conclusion, while AI is not a standalone solution, its integration into breast cancer screening represents an exciting frontier. The UCLA-led findings underscore that AI can identify interval cancers previously missed by radiologists, highlighting the technology’s significance as an adjunct tool. Future studies are essential to validate these results prospectively, optimize AI’s accuracy, and establish best practices for clinical integration. Such efforts promise to transform breast cancer care by facilitating earlier diagnosis, more personalized treatments, and ultimately improved survival rates for patients worldwide.
Subject of Research: Detection of Interval Breast Cancers Using Artificial Intelligence in Mammographic Screening
Article Title: AI-Enhanced Detection of Interval Breast Cancers in U.S. Mammography Screening
Web References:
- Study published in the Journal of the National Cancer Institute: https://academic.oup.com/jnci/advance-article/doi/10.1093/jnci/djaf103/8116029
References:
- Yu, T. et al. Use of Artificial Intelligence for Early Identification of Interval Breast Cancers on Mammograms. Journal of the National Cancer Institute, 2023. DOI: 10.1093/jnci/djaf103
Keywords: Breast cancer, interval cancer, mammography, artificial intelligence, digital breast tomosynthesis, cancer screening, machine learning, radiology, early detection
