Artificial Intelligence Shows Promise in Early Detection of Interval Breast Cancers, UCLA Health Study Finds

A new study led by investigators at the UCLA Health Jonsson Comprehensive Cancer Center suggests that artificial intelligence (AI) could help detect interval breast cancers — those that develop between routine screenings — before they become more advanced and harder to treat. This could potentially lead to better screening practices, earlier treatment and improved patient outcomes. The groundbreaking research, published in the prestigious Journal of the National Cancer Institute, highlights a significant leap forward in diagnostic capabilities, leveraging advanced computational power to augment human expertise in a critical area of cancer care.

The Persistent Challenge of Interval Breast Cancers

Breast cancer remains one of the most common cancers among women worldwide, and early detection is paramount to successful treatment and improved survival rates. Routine mammography screenings have been a cornerstone of this early detection strategy for decades. However, even with advanced screening technologies, a subset of cancers, known as interval breast cancers, still emerges between scheduled mammograms. These cancers are particularly concerning because they tend to be more aggressive, grow faster, and are often diagnosed at a later stage, leading to more complex treatment protocols and, unfortunately, a poorer prognosis compared to screen-detected cancers.

Interval cancers typically account for 20-30% of all breast cancer diagnoses, despite regular screening efforts. They pose a significant challenge to radiologists, who meticulously review thousands of mammograms annually. The subtle signs of these rapidly developing or previously undetected tumors can be incredibly elusive, often missed due to their faint appearance or because they were simply not present or visible at the time of the previous screening. The ability to identify these cancers earlier represents a critical unmet need in breast cancer management, prompting researchers to explore innovative solutions like artificial intelligence.

AI’s Potential: Flagging Subtle Signs

The UCLA Health study found that AI was able to identify "mammographically-visible" types of interval cancers earlier by flagging them at the time of screening. These include tumors that are visible on mammograms but not detected by radiologists, or have very subtle signs on mammography that are easy to miss because the signs were faint or arguably below the level of detection by the human eye. The AI’s strength lies in its capacity to analyze vast amounts of data and recognize patterns that might escape human perception, especially when fatigue or the sheer volume of cases comes into play.

Researchers estimate that incorporating AI into screening could help reduce the number of interval breast cancers by a substantial 30%. This figure is not merely a statistical projection; it translates directly into potentially thousands of women each year receiving earlier diagnoses, benefiting from less invasive treatments, and ultimately experiencing better health outcomes. "This finding is important because these interval cancer types could be caught earlier when the cancer is easier to treat," stated Dr. Tiffany Yu, assistant professor of Radiology at the David Geffen School of Medicine at UCLA and first author of the study. "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." Her statement underscores the profound human impact of such technological advancements.

Pioneering AI Research in the U.S. Context

While similar research exploring the utility of AI in breast cancer detection has been conducted in Europe, the UCLA Health study stands out as one of the first to specifically investigate the use of AI to detect interval breast cancers within the unique framework of screening practices in the United States. This geographical distinction is crucial, as there are key differences between U.S. and European screening paradigms that significantly influence how AI tools might be implemented and their potential efficacy.

In the U.S., the prevailing practice involves annual mammograms, predominantly utilizing digital breast tomosynthesis (DBT), commonly known as 3D mammography. DBT provides a series of thin, high-resolution images of the breast, which can be viewed as a 3D reconstruction, significantly enhancing the detection of small tumors and reducing false positives compared to traditional 2D mammography. Conversely, European screening programs typically employ digital mammography (DM), often referred to as 2D mammography, and adhere to a less frequent screening schedule, usually every two to three years. These variations in technology and screening intervals mean that AI models developed and validated in one context may not be directly transferable or optimally perform in another without specific adaptation and testing. The UCLA study addresses this gap, providing vital data relevant to the U.S. healthcare system.

Methodology: A Deep Dive into Past Screenings

The retrospective study analyzed an extensive dataset comprising nearly 185,000 past mammograms performed between 2010 and 2019. This dataset included both DM and DBT images, offering a comprehensive view of screening practices over nearly a decade. From this vast pool of data, the research team meticulously focused on 148 specific cases where a woman was subsequently diagnosed with interval breast cancer. These cases formed the core of their investigation, allowing them to retrospectively apply AI to the initial screening images that had, at the time, been deemed negative by human radiologists.

A critical step in the study involved a detailed re-review of these 148 interval cancer cases by experienced radiologists. The objective was to determine precisely why the cancer was not spotted earlier. To categorize the interval cancers, the researchers adapted a classification system previously developed and used in Europe. This system provided a standardized framework to understand the nature of the missed diagnoses. The categories included:

• Missed reading error: The cancer was visible on the mammogram but overlooked by the radiologist.
• Minimal signs-actionable: Very subtle signs were present, which, in retrospect, should have prompted further investigation.
• Minimal signs-non-actionable: Extremely faint signs were present, arguably below the threshold for human detection, even with careful review.
• True interval cancer: The cancer developed rapidly in the period between a negative screening and the subsequent diagnosis, meaning it was genuinely not present or visible at the time of the initial mammogram.
• Occult: The cancer was truly invisible on mammogram, detectable only through other modalities like ultrasound or MRI.
• Missed due to a technical error: Issues with image acquisition or processing contributed to the missed diagnosis.

Once the interval cancers were classified, researchers then applied a commercially available AI software called Transpara to the initial screening mammograms performed before the cancer diagnosis. The AI tool’s purpose was to determine if it could detect the subtle signs of cancer that were missed by radiologists during initial screenings, or at least flag them as suspicious. Transpara, a widely recognized AI solution in radiology, scores each mammogram from 1 to 10 for cancer risk. For the purpose of this study, a score of 8 or higher was considered to be flagged as potentially concerning, indicating a high likelihood of malignancy and prompting closer scrutiny.

Key Findings: Promise and Nuance

The application of AI to the retrospective data yielded compelling, albeit nuanced, results. The AI demonstrated a significant ability to identify subtle abnormalities that human radiologists had initially missed. This capability points towards AI’s potential as a powerful adjunct to human expertise, acting as a "second pair of eyes" that could enhance diagnostic accuracy and consistency.

However, the study also provided crucial insights into the current limitations and areas for improvement in AI diagnostic tools. "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," noted Dr. Hannah Milch, assistant professor of Radiology at the David Geffen School of Medicine and senior author of the study. Her statement highlights the responsible scientific approach taken by the UCLA team, acknowledging both the breakthroughs and the challenges.

A particularly striking finding related to occult cancers: "For example, despite being invisible on mammography, the AI tool still flagged 69% of the screening mammograms that had occult cancers. However, when we looked at the specific areas on the images that the AI marked as suspicious, the AI did not do as good of a job and only marked the actual cancer 22% of the time." This observation is profoundly important. While the AI demonstrated an impressive ability to identify that something was wrong in a significant proportion of occult cases (leading to a high overall flagging rate), its precision in localizing the exact area of the cancer within the image was considerably lower. This suggests that while AI can serve as an excellent filter or alert system, its current generation may still require human radiologists to pinpoint the exact location for further investigation, especially for lesions that are not clearly visible. The challenge of interpreting "AI flags" for non-visible lesions introduces complexities for clinical workflow and patient management.

The Broader Landscape of AI in Medical Imaging

The UCLA Health study takes place within a rapidly evolving landscape where artificial intelligence is increasingly integrated into various facets of medicine, particularly in diagnostic imaging. AI algorithms, powered by machine learning and deep learning, are being developed to assist radiologists in detecting abnormalities, quantifying disease progression, and even predicting patient responses to treatment. In mammography specifically, AI has shown promise in improving screening efficiency by triaging studies, reducing reading times, and potentially decreasing false positive rates.

The development of AI tools for medical imaging follows a rigorous path, typically involving:

1. Data Acquisition: Gathering large, diverse datasets of medical images.
2. Algorithm Training: "Teaching" the AI model to recognize patterns associated with specific conditions.
3. Validation: Testing the AI’s performance against human experts and ground truth diagnoses.
4. Clinical Implementation: Integrating validated AI tools into clinical workflows, often as decision support systems.

The current study represents a crucial step in the validation phase, specifically addressing the challenging domain of interval cancers. It builds upon years of research into computer-aided detection (CAD) systems, which have been a precursor to modern AI, but often suffered from high false-positive rates. Today’s AI, with its sophisticated neural networks, offers a much higher degree of accuracy and adaptability.

Implications for Clinical Practice and Patient Outcomes

The findings from the UCLA Health study carry significant implications for the future of breast cancer screening and patient care. If validated in larger, prospective studies, the integration of AI could fundamentally alter screening practices:

• Enhanced Early Detection: The most immediate benefit would be a reduction in the number of advanced interval cancers, leading to diagnoses at earlier, more treatable stages. This directly translates to improved prognoses, less aggressive therapies (e.g., lumpectomy instead of mastectomy, avoidance of chemotherapy), and a better quality of life for patients.
• Radiologist Support: AI could act as an invaluable decision-support tool, helping radiologists manage high caseloads and reduce diagnostic fatigue. By flagging suspicious areas, AI could direct radiologists’ attention to potentially subtle findings they might otherwise overlook. This doesn’t replace the radiologist but augments their capabilities.
• Improved Screening Efficiency: While not explicitly the primary focus of this study, AI has the potential to help prioritize cases, allowing radiologists to focus their expertise on the most complex or concerning mammograms.
• Personalized Screening: In the long term, AI could contribute to more personalized screening approaches, tailoring recommendations based on individual risk factors and the specific characteristics of their mammograms.

Challenges and Future Directions

Despite the exciting promise, the researchers are careful to emphasize that AI is not a panacea and its integration into clinical practice requires careful consideration and further research. "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," Dr. Yu added. "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."

Several key challenges and future research directions emerge from this study:

• Prospective Studies: The current study was retrospective, meaning it looked back at past data. Larger, prospective studies are urgently needed to understand how radiologists would interact with AI in real-time clinical settings. This includes evaluating the impact on radiologist workflow, efficiency, and diagnostic accuracy when AI is actively used during the interpretation process.
• Addressing AI Inaccuracy and Localization: The observation that AI could flag occult cancers but struggled with precise localization is critical. Future AI development needs to focus on improving the spatial accuracy of cancer detection, especially for lesions that are difficult for humans to visualize. How radiologists handle AI flags for areas not visible to the human eye, and when to recommend additional imaging (like ultrasound or MRI) based on AI alerts, needs to be systematically investigated.
• Regulatory Frameworks: As AI tools become more sophisticated, robust regulatory frameworks are essential to ensure their safety, efficacy, and ethical deployment in healthcare. Agencies like the FDA will play a crucial role in evaluating and approving these technologies.
• Integration into Workflow: Seamless integration of AI into existing Picture Archiving and Communication Systems (PACS) and radiology workflows is vital for adoption. The user interface and interpretability of AI outputs will significantly influence its acceptance by clinicians.
• Cost-Effectiveness: The economic implications of integrating AI into screening programs, including the cost of software, hardware, and potential changes in radiologist time, will need to be thoroughly assessed.
• Addressing Bias: Ensuring AI models are trained on diverse datasets to avoid biases that could lead to disparities in care for different demographic groups is paramount.

Expert Perspectives and Broader Impact

The broader scientific and medical communities are keenly watching developments in AI for cancer detection. This UCLA Health study adds significant weight to the growing body of evidence supporting AI’s transformative potential. Experts generally agree that AI will not replace radiologists but rather empower them, allowing them to perform their critical work with enhanced precision and confidence. The ultimate goal remains the same: to improve patient outcomes by catching cancer earlier and enabling more effective treatments.

The work of the UCLA Health Jonsson Comprehensive Cancer Center investigators, including 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, represents a collaborative effort at the forefront of medical innovation. Their research was supported in part by crucial funding from the National Institutes of Health, the National Cancer Institute, the Agency for Healthcare Research and Quality, and Early Diagnostics Inc., underscoring the importance of both public and private investment in advancing diagnostic capabilities. As AI technology continues to mature, studies like this will pave the way for a future where advanced computational tools work hand-in-hand with human expertise to save more lives and alleviate the burden of cancer.