How AI is Transforming Breast Cancer Screening: The Beginning of a New Era in Reducing "Missed Diagnoses"

Breast cancer remains one of the most commonly diagnosed cancers among women worldwide. According to the WHO, approximately 2.3 million people were diagnosed with breast cancer globally in 2022, and about 670,000 died from it. Moreover, about half of the patients are said to have no clear risk factors other than gender and age. This means that even those without a family history or significant lifestyle issues can develop the disease. Therefore, improving the accuracy of early detection continues to be a major challenge for medical fields in various countries.

In response to this challenge, AI is evolving from a "tool to find abnormalities" to a "tool to read future risks." The AI breast cancer screening tool "BRAIx," developed in Australia, is currently gaining attention. According to public reports and research institution announcements, this AI calculates a risk score from 0 to 99.9 based on mammography images and predicts the likelihood of a woman developing breast cancer within the next four years. In studies, about 10% of women who were in the top 2% high-risk category were diagnosed with breast cancer within four years, despite being deemed "normal" at the time of screening. Moreover, this prediction reportedly outperformed traditional indicators such as age, family history, and breast density.

The important point here is that AI does not simply find "tumor-like shadows." Traditional image reading involved determining whether there were clear abnormalities in the current images. However, BRAIx aims to find subtle patterns in the images that the human eye cannot detect and evaluate the "possibility of cancer appearing in the future." The research team explains that by handling information at the pixel level, AI can be effective even in challenging cases like high-density breasts, where distinguishing is difficult. In high-density breasts, both breast tissue and tumors appear whitish, making them prone to being overlooked in traditional mammography. AI is attempting to draw a new auxiliary line in this area.

This trend is not an isolated topic. A large-scale implementation study in Germany reported that AI-assisted double reading increased breast cancer detection rates by 17.6% compared to traditional double reading, without worsening recall rates. Another report conveyed research indicating a reduction in the rate of "interval cancers" found later due to AI utilization. In other words, AI is not a theoretical future technology but is entering the evaluation phase as a "practical tool that can potentially improve screening accuracy."

On the other hand, the attention on AI is also due to the limitations of the screening system itself. In Australia, BreastScreen Australia has contributed to reducing breast cancer mortality rates over the years, but participation rates are still not high enough. Additionally, as a system, screenings are generally conducted at "uniform intervals" for certain age groups, and there is no widespread mechanism to flexibly change screening frequency based on individual future risks. Research teams and related experts emphasize that using AI could bring us closer to personalized screenings, where "high-risk individuals are screened more frequently" and "low-risk individuals face less burden."

This "personalization" is not just convenient for patients. Reducing missed cases in high-risk individuals can lead to earlier treatment initiation. Conversely, avoiding unnecessary additional tests for low-risk individuals can reduce psychological burdens and medical costs. The true value of AI research lies not in replacing doctors but in visualizing where limited medical resources should be prioritized. As the demand for image diagnosis increases with aging populations, the burden on radiologists is rising in many countries. AI is expected not as a "magic solution to manpower shortages" but as an "auxiliary device to focus expert judgment on more important areas."

In fact, the welcoming mood is strong on public social media. On LinkedIn, posts related to cancer research and medical innovation have received positive comments such as "important work," "a big step forward," and "hope it gets implemented in medical settings." Posts from the VCCC Alliance received multiple comments praising the research, and related researchers and medical professionals reacted by calling it "a significant achievement." In AIML-related posts, the narrative that "AI functions as a second eye rather than replacing doctors" is prominent. The enthusiasm on social media indicates that this topic is being perceived not just as a technological news item but as an issue closely tied to urgent challenges in medical settings.

However, the reactions are not entirely laudatory. Expert comments and related reports consistently maintain a cautious stance, stating that "while promising, further validation is needed before incorporating it into routine clinical practice." Expert opinions introduced by Spain's Science Media Centre noted that while the research is large-scale and methodologically strong, comparisons with BRCA mutation carriers should be read as "risk limited to a four-year period" and should not be equated with lifetime risk. Other reports also caution against generalizing results from single facilities or specific groups. Even if AI is highly accurate, it does not necessarily function the same way across all regions, races, and systems.

Furthermore, it is important to remember that AI does not assume the "final responsibility for diagnosis." ABC's report also states that the research side clearly indicates the need for final oversight by human radiologists. In reality, contexts such as patient condition, medical history, palpation findings, patient anxiety, decisions on additional tests, and connection to treatment cannot be completed with images alone. Recently, there have been warnings about relying too much on AI for medical consultations, and in diseases like breast cancer, where delays in response can be fatal, caution is especially required when using AI to supplement self-judgment. AI is smart, but it is still humans who see patients. This fundamental principle does not change even as technology advances.

Nevertheless, the significance of the research presented this time is substantial. Breast cancer screening has long been operated under the design of "something to be done at regular intervals once a certain age is reached." However, in the future, the idea that "even within the same age group, the same frequency may not be appropriate" or "if someone in their 40s is high-risk, earlier intervention may be necessary" might become mainstream. The research team envisions a future screening system based not only on age criteria but also on individual risk. If implementation progresses, breast cancer screening could shift from a "system drawing lines based on age" to a "system tailored to each individual using images and data."

What is truly frightening in breast cancer care is not just the visible abnormalities themselves. It is the gap of time when the disease progresses after being told "it's okay" during screening. AI aims to change precisely this gap. While it is difficult to eliminate all oversights, knowing the risk of approaching an oversight earlier and taking the next step sooner is crucial. In this sense, AI is beginning to show its true potential not as a "machine to find cancer" but as a "predictive device to reduce being too late."

Source URL

https://www.theage.com.au/lifestyle/health-and-wellness/artificial-intelligence-revolutionising-breast-cancer-research-20260307-p5o8cx.html

・Related reports covering the same content (Overview of BRAIx, risk prediction within four years, results of the top 2%, researcher comments)
https://www.abc.net.au/news/2026-03-04/artificial-intelligence-breast-cancer-risk-mammogram/106411246

・Online first publication information in academic journals (Paper title, publication date, formal positioning of the research)
https://www.thelancet.com/journals/landig/onlinefirst

・Announcement by research institutions (Background of BRAIx development, learning of about 400,000 people, verification of about 96,000 people, external confirmation in Sweden, concept of personalized screening)
https://www.svi.edu.au/news-events/ai-based-tool-estimates-breast-cancer-risk-within-next-four-years/

・Summary of scientific information for media (Key points of the research, 9.7% four-year risk in the top 2%, need for additional research before routine care)
https://www.scimex.org/newsfeed/ai-could-help-preduct-your-risk-of-breast-cancer-in-the-next-4-years

・WHO's breast cancer fact sheet (Global number of patients and deaths, risk factors, importance of early detection)
https://www.who.int/news-room/fact-sheets/detail/breast-cancer

・Large-scale implementation study in Germany (Case where AI-assisted reading increased detection rates without worsening recall rates)
https://www.nature.com/articles/s41591-024-03408-6

・Reports supplementing cautious views on AI introduction (While AI support may reduce later discoveries, further validation is needed)
https://www.theguardian.com/science/2026/jan/29/ai-use-in-breast-cancer-screening-cuts-rate-of-later-diagnosis-by-12-study-finds

・Basic information on Australia's screening system (Explanation of BreastScreen Australia, concept of target age group)
https://www.aihw.gov.au/reports/cancer-screening/breastscreen-australia-monitoring-report-2025/contents/breast-cancer/breast-cancer-screening

・Source of public SNS reactions 1 (VCCC Alliance's LinkedIn post and comments)
https://www.linkedin.com/posts/vcccalliance_an-artificial-intelligence-tool-to-predict-activity-7435097386366054400-xWmT

・Source of public SNS reactions 2 (AIML/related posts. Context of AI as a "second eye")
https://www.linkedin.com/posts/jonathangovette_sweden-just-proved-ai-can-catch-29-more-activity-7426652456484397056-ceWM

・Source of public SNS reactions 3 (Research introduction LinkedIn post)
https://www.linkedin.com/posts/professor-gerald-lip-633a5556_ai-based-braix-risk-score-for-the-intermediate-term-activity-7435008570406117376-ypfK