By Billy Candra 
The groundbreaking research, recently published in the prestigious Journal of the National Cancer Institute, offers a compelling vision for the future of breast cancer detection, particularly for those elusive cases that slip through the conventional screening net. Interval breast cancers, by their very nature, pose a significant challenge to clinicians and patients alike. They are defined as cancers diagnosed within a specific interval after a negative mammogram, typically before the next scheduled screening. These cancers often present as more aggressive, grow rapidly, and are associated with a poorer prognosis compared to screen-detected cancers, underscoring the urgent need for enhanced detection methodologies.
The Elusive Threat: Understanding Interval Breast Cancers
Breast cancer remains one of the most prevalent cancers globally, and the second leading cause of cancer death among women in the United States. Regular mammography screening has been instrumental in reducing breast cancer mortality by detecting tumors at earlier, more treatable stages. However, the system is not without its limitations, and interval cancers represent a critical gap.
Approximately 20% to 30% of all breast cancers are diagnosed as interval cancers. These can be particularly distressing for patients who have recently received a "clear" mammogram. The reasons for their missed detection are varied: some are genuinely new growths that emerge rapidly, while others were present at the time of screening but were either too subtle to be detected by the human eye, obscured by dense breast tissue, or misinterpreted by the interpreting radiologist. The latter category, often termed "missed" or "subtle" interval cancers, is precisely where AI holds immense promise.
When an interval cancer is diagnosed, it often means the disease has had more time to progress, potentially leading to larger tumor sizes, lymph node involvement, and a higher likelihood of requiring more aggressive treatments, such as extensive surgery, chemotherapy, and radiation. This stark difference in prognosis highlights why detecting these cancers earlier, ideally at the initial screening stage, could revolutionize patient care and significantly improve survival rates.
UCLA’s Pioneering Study: Methodology and Key Insights
The UCLA Health Jonsson Comprehensive Cancer Center study stands out as one of the first in the United States to rigorously investigate the application of AI in identifying interval breast cancers, especially within the context of U.S. screening practices. While similar research has been conducted in European settings, key differences in screening protocols between the continents make the U.S.-specific findings particularly vital. In the U.S., digital breast tomosynthesis (DBT), commonly known as 3D mammography, is widely used, and annual screenings are typical. European programs, in contrast, more often utilize digital mammography (DM or 2D mammography) and adhere to less frequent screening intervals, usually every two to three years. These variations underscore the importance of validating AI’s efficacy within the distinct U.S. clinical environment.
The retrospective study meticulously analyzed a vast dataset comprising nearly 185,000 past mammograms performed between 2010 and 2019. From this extensive pool, researchers focused on 148 cases where women were subsequently diagnosed with interval breast cancer. A team of experienced radiologists then conducted a thorough review of these 148 cases, aiming to pinpoint why the cancers were not detected during the initial screening.
To standardize their analysis, the research team adapted a European classification system for interval cancers. This comprehensive system categorizes missed cancers into several types:
- Missed reading error: The cancer was visible on the mammogram but overlooked by the radiologist.
- Minimal signs-actionable: Subtle signs of cancer were present and, in hindsight, warranted further investigation.
- Minimal signs-non-actionable: Very faint signs were present, arguably below the threshold for human detection or action.
- True interval cancer: The cancer was genuinely not present or visible on the prior mammogram and developed rapidly thereafter.
- Occult: The cancer was truly invisible on the mammogram, regardless of interpretation.
- Missed due to a technical error: Issues with image acquisition or processing led to the missed detection.
This detailed categorization provided a robust framework for understanding the nature of missed diagnoses and for evaluating the AI’s potential contribution.
Following the human expert review, the researchers applied a commercially available AI software, Transpara, to the initial screening mammograms that preceded the interval cancer diagnoses. The AI tool’s objective was to determine if it could identify subtle cancerous signs that were either missed by radiologists or flag suspicious areas for further scrutiny. Transpara operates by assigning a risk score from 1 to 10 for cancer presence on each mammogram, with a score of 8 or higher triggering a "potentially concerning" flag.
The AI’s Performance: Promise and Pitfalls
The study’s findings revealed a dual narrative of significant promise alongside crucial areas for further refinement. The most striking finding was the researchers’ estimation that incorporating AI into routine screening protocols could lead to a substantial 30% reduction in the number of interval breast cancers. This figure represents a dramatic improvement in early detection capabilities, particularly for the "mammographically-visible" types of interval cancers – those that are present on the image but either too subtle or overlooked by human interpretation.
Dr. Tiffany Yu, assistant professor of Radiology at the David Geffen School of Medicine at UCLA and the study’s first author, emphasized the profound impact of this potential: "This finding is important because these interval cancer types could be caught earlier when the cancer is easier to treat. For patients, catching cancer early can make all the difference. It can lead to less aggressive treatment and improve the chances of a better outcome." This sentiment resonates deeply with the core objective of cancer screening: to intervene before the disease escalates, thereby preserving quality of life and extending survival.
However, the study also provided a candid look at the current limitations of AI. Dr. Hannah Milch, assistant professor of Radiology at the David Geffen School of Medicine and senior author of the study, highlighted these complexities: "While we had some exciting results, we also uncovered a lot of AI inaccuracy and issues that need to be further explored in real-world settings."
One notable example of this inaccuracy pertained to occult cancers – those truly invisible on mammography. Despite their non-visibility, the AI tool surprisingly flagged 69% of the screening mammograms associated with occult cancers as suspicious. While this might initially seem positive, a deeper dive revealed a critical flaw: when examining the specific areas on the images that the AI marked as suspicious, the tool accurately pinpointed the actual cancer location only 22% of the time. This suggests that while AI can sometimes detect general patterns of concern, its ability to precisely localize truly invisible cancers remains a significant hurdle. Such imprecision could lead to an increase in false positives, unnecessary patient recalls, and potentially biopsy procedures, which carry their own risks and anxieties.
The Broader Implications for Clinical Practice
The integration of AI into breast cancer screening holds immense potential to reshape clinical practice, offering a "second set of eyes" for radiologists. The sheer volume of mammograms processed daily can lead to fatigue, and subtle abnormalities are often challenging to identify, especially in dense breast tissue which can obscure cancerous lesions. AI, with its tireless analytical capacity, could serve as an invaluable aid, flagging suspicious areas that human radiologists might miss.
This paradigm shift could lead to several benefits:
- Earlier Diagnosis: Reducing interval cancers means more patients receive a diagnosis at an earlier stage, when treatment options are more numerous and less invasive.
- Improved Patient Outcomes: Early detection is directly linked to higher survival rates and better long-term prognoses. For instance, the 5-year survival rate for localized breast cancer is over 99%, dropping significantly for regional or distant disease.
- Optimized Radiologist Workflow: AI could help prioritize cases, bringing potentially challenging or high-risk mammograms to the forefront for immediate and meticulous review, thereby enhancing efficiency.
- Reduced Anxiety: For patients, knowing that advanced AI technology is assisting in their screening could offer an added layer of reassurance, though the potential for increased false positives also needs careful management.
Challenges and Future Directions: A Path Forward
Despite the promising outcomes, the UCLA study underscores that AI is not a standalone solution. Its integration into clinical practice must be approached thoughtfully, addressing the identified inaccuracies and logistical challenges.
A critical next step involves conducting larger, prospective studies. Retrospective studies, while valuable for identifying patterns, cannot fully replicate the dynamic environment of a real-world clinic. Prospective research would observe how radiologists actually utilize AI in real-time, how their workflow is impacted, and how patient outcomes are affected. Key questions such as how to manage cases where AI flags suspicious areas not visible to the human eye, or when the AI’s localization of cancer is inaccurate, must be thoroughly explored.
Ethical considerations also loom large. The responsibility for diagnosis ultimately rests with the human clinician. Clear guidelines will be needed on how to interpret and act upon AI’s recommendations, especially when they diverge from human assessment. Patient communication regarding the role of AI in their screening will also be crucial to ensure transparency and trust.
Dr. Yu summarized the nuanced perspective: "While AI isn’t perfect and shouldn’t be used on its own, these findings support the idea that AI could help shift interval breast cancers toward mostly true interval cancers. It shows potential to serve as a valuable second set of eyes, especially for the types of cancers that are the hardest to catch early. This is about giving radiologists better tools and giving patients the best chance at catching cancer early, which could lead to more lives saved." This perspective positions AI not as a replacement for human expertise, but as a powerful augmentation.
Broader Context: The Ascent of AI in Healthcare
The UCLA study is part of a larger, accelerating trend of AI adoption across various medical disciplines. From drug discovery and personalized medicine to diagnostic imaging and surgical assistance, AI is poised to revolutionize healthcare delivery. In radiology specifically, AI algorithms are being developed for detecting a wide array of conditions, from lung nodules and brain anomalies to diabetic retinopathy. The promise lies in AI’s ability to process vast amounts of data, identify subtle patterns, and potentially reduce diagnostic errors, thereby enhancing the precision and efficiency of medical care.
However, the path to widespread implementation requires rigorous validation, regulatory oversight, and careful consideration of the human-AI interface. The success of AI in improving patient outcomes will ultimately depend on its seamless and responsible integration into the existing healthcare ecosystem, with the human clinician remaining at the center of patient care.
The UCLA Health Jonsson Comprehensive Cancer Center’s research provides a significant step forward in this journey, illuminating a viable pathway to mitigate the challenging problem of interval breast cancers and ultimately, to save more lives.
Other authors involved in this pivotal research, all from UCLA, include Dr. Anne Hoyt, Dr. Melissa Joines, Dr. Cheryce Fischer, Dr. Nazanin Yaghmai, Dr. James Chalfant, Dr. Lucy Chow, Dr. Shabnam Mortazavi, Christopher Sears, Dr. James Sayre, Dr. Joann Elmore and Dr. William Hsu. The work received vital support in part from the National Institutes of Health, the National Cancer Institute, the Agency for Healthcare Research and Quality, and Early Diagnostics Inc., highlighting the collaborative and multi-faceted effort required for such impactful scientific endeavors.