Breaking Through the Breast Cancer Screening Bottleneck: How AI is Transforming Early Detection for Over 300,000 Patients

At present, public awareness regarding the prevention and screening of breast cancer remains largely vague.

In early 2021, a medical aesthetics clinic catering to women offered its members a special “benefit”:Under the organization’s leadership, eligible members may visit partner hospitals for free breast cancer screening during designated periods.Ultimately, approximately 200 women participated in this event. Nearly 20 of them were identified as being at risk for breast cancer based on examination results and were notified to undergo further diagnostic evaluation. The fact that multiple women with potential breast cancer risks were detected during an unintentional screening initiative by a medical aesthetics institution underscores the poor penetration of breast cancer screening programs. On a broader societal level, the implementation and promotion of early screening and diagnosis for breast cancer will face even more formidable challenges.

In fact, China had already launched a national program in 2009 to provide free screening for cervical and breast cancer (“two cancers”) among rural women. To promote early diagnosis and treatment, the target population for this screening was gradually expanded from eligible rural women to include eligible urban and rural women across the country in 2022, thereby progressively improving women’s health outcomes.

According to the latest cancer statistics, there were 2.26 million new cases of breast cancer, officially surpassing lung cancer to become the most common cancer worldwide. As part of these statistics, China had approximately 420,000 newly diagnosed breast cancer patients in 2020, and this number continues to rise year by year.

While the number of diagnosed patients continues to rise year by year, the medical community has never halted its research efforts in screening, diagnosis, and treatment. Under the current healthcare system, there is a high degree of consensus among physicians regarding the diagnosis and treatment of breast cancer. If detected early, the total treatment cost amounts to only a few thousand yuan; however, the key lies inWe cannot accurately screen every patient with early-stage breast cancer.

Population-wide breast cancer screening under the current mechanism is primarily conducted throughInitial Screening InstitutionandReceiving InstitutionTwo major institutions collaborate on implementation.

Initial Screening InstitutionTypically composed of primary healthcare institutions, this approach employs ultrasound examination for breast cancer screening in patients and assigns BI-RADS categories. Categories 1–2 require regular follow-up; categories 0 and 3 warrant further evaluation with mammography; and categories 4–5 are recommended for biopsy and subsequent histopathological examination.

Although the BI-RADS classification system is sufficiently objective, physicians’ assessments are not always accurate. In actual clinical ultrasound practice, there is a significant shortage of sonographers in primary healthcare institutions. Coupled with high workloads and susceptibility to fatigue, it is difficult for physicians to maintain sustained concentration, which increases the risk of missed or misdiagnoses. Furthermore, there is considerable inter-observer variability in ultrasound diagnosis and BI-RADS categorization among different physicians. Inaccurate classification and diagnosis can easily delay patient treatment, severely undermining the utilization efficiency of medical resources and the credibility of primary healthcare institutions.

RevisitingMedical InstitutionPatients classified as BI-RADS category 0 or 3 need to return to the medical institution for follow-up visits and undergo repeat mammography. In European and American countries, mammography is commonly used for initial breast cancer screening, as it more effectively detects early-stage lesions and microcalcifications. However, due to the dense breast tissue characteristic of Asian women, mammograms often produce significant overlapping shadows, requiring highly experienced technicians and physicians for image acquisition and interpretation. Currently, the prevalence of mammography in China is lower than that of ultrasound, primarily for two reasons: First, image acquisition and interpretation are performed by technicians and physicians at different stages; when radiologists find that the images do not meet diagnostic requirements, patients have usually already left the hospital, making quality control of mammographic imaging a constraint on its development. Second, there are currently only about a thousand physicians in China specializing in mammographic diagnosis, with inter-reader variability among physicians of different experience levels reaching up to 30%, and a severe shortage of interpretive talent in primary care hospitals.

Attempting to meet massive diagnostic demands with scarce medical resources often leads to various issues. A company specializing in the research and development of AI-assisted diagnostic technology for breast medical imaging told VCBeat, “When developing our algorithms, we had hoped to identify some typical cases from over 400 hospital cases. However, after review, we found that only 10%–20% of the mammographic images met quality standards.”

Despite numerous challenges in the screening and prevention of breast cancer, the core issues ultimately revolve around two key elements: “healthcare supply” and “quality of care.” Wang Shui, a member of the National Committee of the Chinese People’s Political Consultative Conference, Vice President of Jiangsu Province Hospital, and Director of the Breast Disease Diagnosis and Treatment Center, also stated at a symposium in February on “Standardizing Clinical Oncology Drug Use to Ensure Medical Quality and Safety” that while the demand for breast cancer screening among Chinese women is gradually increasing, the overall quality of screening remains insufficient. He therefore recommended integrating new technologies such as artificial intelligence (AI) into cancer screening to improve screening quality.

Current mainstream AI companies typically address breast cancer screening and prevention by focusing on “supply” and “quality,” leveraging AI for assisted quality control and assisted diagnosis.

Assisted Quality Control, AI is leveraged to provide quality control during mammography acquisition by physicians, thereby optimizing the diagnostic process. In traditional workflows, technologists cannot assess image quality in real time after acquisition; issues are often identified only when radiologists review the images for diagnosis, revealing that some films are unsuitable for diagnostic purposes. At this stage, recalling patients for repeat examinations imposes a burden on both patients and healthcare providers. In this context, implementing AI-driven real-time quality control for mammography can reduce ineffective examinations while enhancing image quality. When integrated with computer-aided diagnostic tools, some enterprises have achieved detection rates exceeding 90%.

Auxiliary Diagnosis, lesion detection, analysis, and BI-RADS categorization during mammography are critical components of the examination. In terms of lesion detection and analysis, AI can rapidly and accurately determine lesion types, perform information analysis, and delineate suspicious lesions with bounding boxes, thereby assisting physicians in lesion identification and analysis. Regarding BI-RADS categorization, unlike the subjective assessments often made by primary-care physicians, AI leverages breast cancer diagnostic criteria combined with high-quality annotated data and employs technologies such as artificial intelligence and deep learning to deliver more accurate and standardized categorization results.

In summary, for mammographic diagnosis, high-quality imaging and precise, efficient detection and analysis are crucial. This places higher demands on AI capabilities; furthermore, simultaneously meeting the comprehensive needs spanning from image acquisition to detection and diagnosis, clinical treatment, and even research and training is exceptionally challenging. While most AI medical imaging companies are expanding “horizontally” by covering a broader range of diseases, MEDICAL AI has concentrated its R&D efforts to develop “vertically,” focusing specifically on breast imaging. It has achieved an integrated, full-stack intelligent solution for breast health encompassing “screening, diagnosis, treatment, research, and education.” Additionally, it has launched the “Pink Care AI” campaign across multiple provinces in China to promote women’s breast health.

Revisiting Ultrasound Imaging: Currently, the number of enterprises engaged in AI-driven ultrasound research is relatively small compared to conventional AI applications such as pulmonary nodule detection and fundus imaging, and the adoption rate of related products remains comparatively slow. This is primarily attributable to the inherently high technical barriers associated with artificial intelligence in ultrasound. Ultrasound data incorporates an additional temporal dimension beyond the two-dimensional data generated by mammography. This implies that while image acquisition and diagnostic interpretation in mammography can be performed with ample time, such luxury is absent in ultrasound diagnostics, where image acquisition and diagnosis must be conducted simultaneously.

Currently, very few companies worldwide possess the capability to develop real-time dynamic algorithms for ultrasound. MEDICAL AI has broken through the limitations of static 2D imaging, pioneering the field of dynamic real-time lesion detection. Its independently developed AI-assisted diagnostic model for ultrasound equipment employs Neural Architecture Search (NAS), achieving a processing speed of 64 frames per second with a detection latency of less than 0.09 seconds. In other words, while physicians perform ultrasound scans, the AI algorithm can accurately capture lesions that flash by in merely milliseconds, enabling real-time lesion detection and analysis along with BI-RADS classification recommendations, truly realizing "what you see is what is diagnosed."

Compared with conventional AI ultrasound products, MEDICAL AI’s real-time dynamic intelligent analysis system for breast ultrasound excels in"Dynamic Real-Time Monitoring"。

According to Cao Ying, Vice President of MEDICAL AI, traditional artificial intelligence algorithms cannot meet the detection requirements for ultrasound scans due to their dynamic and continuous nature. Therefore, ultrasound AI must be equipped with dynamic and real-time detection technologies and capabilities to deliver clinical value. However, from the perspective of technological research and development, challenges such as the lack of publicly available dynamic datasets and related algorithmic research persist. The R&D team at MEDICAL AI overcame numerous obstacles by independently collecting data to establish a proprietary dynamic database and achieving breakthroughs in dynamic algorithms, thereby realizing truly “dynamic and real-time” lesion detection and analysis. This achievement represents not only a technological breakthrough for MEDICAL AI but also an innovation in the field of medical imaging, marking a solid step forward toward becoming a specialized, refined, distinctive, and innovative enterprise.

In May 2021, MEDICAL AI completed clinical validation on more than 2,000 cases at top-tier tertiary hospitals in China. The data showed that its AI achieved a detection rate of up to 98% for breast lesions and an accuracy rate of up to 90% in determining the benign or malignant nature of tumors. Experts commented that “these metrics are comparable to those of senior physicians at top-tier tertiary hospitals.”

In August 2021, the State Council issued the Outline for the Development of Chinese Women (2021–2030), requiring the gradual implementation over the next decade of a comprehensive prevention and control system and assistance policies for cervical cancer and breast cancer within basic public health services. The goal is to achieve a resident awareness rate of over 90% for cervical and breast cancer prevention and control knowledge; to reach a population screening rate of over 70% for cervical cancer among eligible women; and to steadily increase the population screening rate for breast cancer.

A researcher in breast medical imaging told VCBeat that the current screening rate for breast cancer among age-eligible women is approximately 20%–30%, and it will be difficult to achieve a screening level comparable to that of cervical cancer without leveraging artificial intelligence.

Although ultrasound and mammography can address the basic screening needs for breast cancer, a tumor biopsy is still required to make a definitive diagnosis for more accurate determination of lesion benignity or malignancy.

The core issue lies in interdisciplinary collaboration. In breast surgery, the use of ultrasound can improve surgical outcomes, significantly reducing tumor residual rates, re-excision rates, and the volume of excised specimens. However, ultrasound guidance presents a technical barrier for surgeons, necessitating collaboration with sonographers. In practice, sonographers are a scarce resource; in megacities such as Shanghai and Beijing, scheduling an ultrasound-guided breast tumor biopsy can take nearly a month.

The “MEDICAL AI Real-time Dynamic Intelligent Analysis System for Breast Ultrasound” mentioned above can also address this issue. Leveraging artificial intelligence technology, the system enables precise localization of lesions and intelligently generates surgical pathways, guiding surgeons in delivering precise treatment, shortening operative time, reducing surgical risks, and improving the success rate of surgical interventions.

Another advantage of this system lies in optimizing physician efficiency. Previously, physicians had to periodically pause during ultrasound examinations to capture screenshots for documentation and complete records after the procedure. With AI support, these additional tasks during the procedure will be automatically performed by artificial intelligence, allowing physicians to devote more attention to diagnosis and treatment itself.

During this year’s Two Sessions, Yu Jinming, a deputy to the National People’s Congress (NPC), academician of the Chinese Academy of Engineering, and president of Shandong Cancer Hospital, along with NPC deputy Wang Ling, both proposed including screening for gynecological cancers such as cervical cancer and breast cancer in the national medical insurance coverage. Deputy Wang Ling further suggested, “Currently, breast cancer screening relies on manual ultrasound and mammography, which demand high technical standards for both equipment and personnel, making comprehensive implementation in rural areas challenging. We hope the state will increase investment in artificial intelligence for breast cancer screening.”

Across the industry, multiple companies have already developed mature, high-quality AI-based breast cancer screening technologies and are in the process of deploying them in tertiary hospitals and primary care settings. However, it always takes time for a cutting-edge technology to move from the laboratory to every corner of clinical practice.

The Ripple Effects of AI Penetration in Breast Care Are Emerging, Accelerating the Adoption of Intelligent Breast Cancer Diagnosis. An increasing number of medical institutions are deploying AI-powered breast diagnostic tools, enabling more patients to receive early treatment and significantly reducing their financial burden.

However, it is crucial to recognize that for breast cancer screening to become a nationwide program, its ultimate implementation must rest at the primary care level. Yet, patient reluctance to seek care at primary healthcare institutions persists. In the absence of adequate payment guarantees and effective resource allocation, the widespread deployment of breast AI in primary care settings will not be realized anytime soon. Policy support is needed as a safety net, along with greater participation from enterprises and medical institutions, to accelerate the adoption of breast AI technologies.