Abstract
Introduction
Breast cancer Mammography (MAM) screening was proven to improve survival worldwide. However, younger patients with higher breast density made MAM less effective in China. It is necessary to establish Chinese-specific effective screening strategies. This study aims to explore the efficacy of artificial intelligence (AI)-assisted ultrasound breast cancer screening in China.
Methods
Eligible participants were those aged 35–69 years and were attending the Chinese "Two Cancer (breast and cervical cancer) Screening" program. Two districts were selected as cluster to receive either AI-assisted ultrasound screening or routine ultrasound screening. We obtained data on cancer diagnosis through active follow-up and linkage with municipal cancer registry. The primary outcome was improved screening sensitivity enabling the detection of more true-positive cases. This study is registered at ClinicalTrials.gov under the number NCT06521788 (Initial Release Date: 07/22/2024).
Results
A total of 21,790 individuals in two districts were included in this study, with 8,736 participants in Hongkou district receiving AI-assisted ultrasound screening and 13,054 in Pudong district undergoing routine ultrasound screening. Of the 21,790 screened participants, 232 (10.7‰) tested positive, with AI detecting similar positivity rates compared to routine screening (12.2‰ vs. 9.6‰, P = 0.07). After one year of follow-up, 49 participants were diagnosed with breast cancer: 30 were screen-detected cancers, and 19 were interval cancers. The AI group demonstrated a significantly higher screening sensitivity (75%, 95% CI 54.8–88.6) compared to the routine group (42.8%, 95% CI 22.6–65.6). AI-assisted screening identified more breast cancers than the routine screening group (AI: 21 of 8736; routine: 9 of 13,054, P = 0.001). However, there was no significant difference between the two groups in terms of interval cancer detection (AI: 7 of 8736; routine: 12 of 13,054, P = 0.789). Furthermore, the proportion of early-stage cancers among screen-detected cases was significantly higher in the AI group (95.2%, 20/21) than in the routine group (88.9%, 8/9; p < 0.001).
Conclusions
AI-assisted ultrasound screening significantly increases the detection rate of early breast cancers.
Trial registration This study is registered at ClinicalTrials.gov under the number NCT 06521788 (Initial Release Date: 07/22/2024).
Keyword: Breast cancer screening, Artificial intelligence, Breast ultrasound
Introduction
Breast cancer is the most common malignant tumor among women worldwide [1]. In China, it also ranks as the most prevalent malignant tumor among women, with its incidence exhibiting a rapid upward trend in recent years [2].
Researches have demonstrated that large-scale population-based screening can significantly reduce breast cancer mortality [3–10]. In the United States, breast cancer mortality declined from an age-adjusted rate of 48 deaths per 100,000 women in 1975 to 27 per 100,000 women in 2019. This decline was largely attributed to widespread screening, with a coverage rate of nearly 70% [10], along with advances in treatment, better drug therapies, and more efficient diagnostic and treatment pathways. In 2013, the World Health Organization (WHO) released a position paper on mammography screening and guidelines for the referral of suspected breast cancer cases in primary healthcare settings in low-resource environments, recommending that countries integrate breast cancer screening and early detection programs into their primary healthcare systems [11].
Mammography (MAM) is widely employed in breast cancer screening in European and American countries. However, its application among Asian women, including those in China, presents significant limitations. MAM is more effective for screening older women or those with lower breast density, but it has reduced sensitivity for younger women and those with higher breast density [9]. Notably, Asian women exhibit significantly higher breast density compared to their counterparts in Europe and America. In China, the proportion of women with dense breasts [Breast Imaging Reporting and Data System (BI-RADS) categories C/D] is as high as 70.2% [12], whereas the American College of Radiology reports a proportion of about 50% in the U.S. population [13]. Additionally, Chinese women tend to develop breast cancer at a younger age [2].
Breast ultrasound can address the limitations of mammography and is a more suitable screening method for Chinese women. Studies have shown that the cancer detection rate with ultrasound is comparable to that of mammography, with a higher proportion of invasive cancers detected [14–17]. Additionally, ultrasound demonstrates better cost-effectiveness. The latest breast cancer screening and early diagnosis guidelines in China recommended ultrasound screening in young women and those with dense breast tissues [18].
In recent years, artificial intelligence (AI) has been increasingly applied in the medical field [19–22]. AI-assisted ultrasound technology can further enhance the efficacy of breast cancer screening. The Fudan University Shanghai Cancer Center (FUSCC) has independently developed a portable AI-assisted ultrasound diagnostic device, utilizing machine learning algorithms based on data from over 300,000 female breast lesions [23–25]. Previous studies have confirmed that this AI-assisted ultrasound demonstrates strong identification capabilities for breast lesions in Chinese women. Its ability to recognize high-risk breast lesions (BI-RADS 4A and above) and early breast cancer is comparable to that of routine ultrasound, making it suitable for screening in large-scale community settings with women at general risk [26]. We subsequently conducted a prospective study within a community population to evaluate the effectiveness of AI-assisted ultrasound in breast cancer screening.
Methods
Study population
China has launched the “Two Cancers (Breast and Cervical Cancer) Screening” project since 2009, providing free screening for millions of rural women aged 35–69 years. And this cancer screening program has expanded to include all eligible women in both rural and urban areas as a basic public health service since 2019 [27, 28].
Building on this screening initiative, we conducted a prospective controlled trial in Shanghai, China. Eligible participants were women aged 35–69 years, who were attending the “Two Cancers (Breast and Cervical Cancer) Screening” project, and had no history of breast cancer (including in-situ cancer), or any other cancers in the previous 5 years. Participants were also required to have no serious cardiopulmonary insufficiency, liver or kidney insufficiency, or other systemic diseases, and a life expectancy of more than five years (Fig. 1).
Fig. 1.
Trial profile
This study was conducted in accordance with the guidelines of the Helsinki Declaration and was approved by the Institutional Review Board of Fudan University Shanghai Cancer Center (No. 2008223-22). Informed consent was obtained from all individual participants included in the study.
This study is registered at ClinicalTrials.gov under the number NCT 06521788 (Initial Release Date: 07/22/2024).
Definition and selection of a cluster
We define one district as a cluster. As of 2021, Shanghai has jurisdiction over 16 municipal districts. The division of these districts was based on field-measured data and relevant geographical maps, utilizing human–computer interaction for the vectorization of administrative division maps. Among these divisions, seven are classified as urban areas and nine as suburban areas. The urban districts cover an area of 20–40 square kilometers, with a resident population ranging from 600,000 to 1,000,000. In contrast, the suburban districts span 300–1200 square kilometers and have populations between 500,000 and 3,200,000. Each district has at least 20,000 women aged 35–69.
We selected two districts as the units of cluster rather than individuals based on economy, population characteristics, pandemic, and willingness to cooperate. Hongkou district was selected to serve as the intervention group and Pudong district as the control group.
Trial participants and intervention
Women in the intervention group received AI-assisted ultrasound screening, while those in the control group underwent routine ultrasound screening. The AI-assisted ultrasound diagnostic device used in this study is a portable and intelligent AI-assisted ultrasound diagnostic instrument developed collaboratively by FUSCC, the School of Information Science and Engineering at Fudan University, Shanghai University, and Shisun Intelligent Technology (Shanghai) Co., Ltd. This project was funded by the Major Instruments Program of the National Natural Science Foundation of China and the Science and Technology Innovation Action Plan of the Shanghai Municipal Science and Technology Commission. The AI-assisted device features a portable design, consisting of a display panel measuring 500 mm × 500 mm × 20 mm laptop and an ultrasound probe (frequency of 10 MHz, 4 cm in width), which can be easily carried in a bag for on-the-go examinations, making it ideal for community/rural screening (Fig. 2). During the ultrasound scanning process, it enables real-time breast nodule lesion detection. When suspicious lesions are detected, the system issues both audio and visual alerts, with the suspicious lesions outlined by a red bounding box. After lesion localization, the system provides a detailed description of the lesion characteristics, including morphology, boundary, margin, echogenicity, as well as the detection and analysis of associated features. Based on the BI-RADS classification guideline, the AI system conducts a comprehensive evaluation of the lesion and assigns a BI-RADS category. The routine ultrasound device was the IU22 ultrasound diagnostic system from Philips, Netherlands, with the L9-3 probe and a frequency range of 3–9 MHz.
Fig. 2.
Ultrasound doctors used portable AI-assisted ultrasound diagnostic instrument for screening on-site
Participants in both groups were eligible for further diagnostic evaluation and treatment at Fudan University Shanghai Cancer Center. Recruitment began in January 2021 and was completed in December 2022. All participants were followed up until the end of 2023, and the database was locked in January 2024 for analysis.
Screening procedure
Our screening procedure is based on the National Breast Cancer Screening Process Technical Guidelines, issued in 2015 by the National Health Commission of the People’s Republic of China [27]. The procedure consists of the following steps (Fig. 3):
Clinical examination and initial breast ultrasound screening: All participants were enrolled and underwent a clinical examination along with an initial breast ultrasound (routine ultrasound or AI-assisted ultrasound). The results of the initial screening tests were classified according to the BI-RADS.
MAM Rescreen: Participants who received BI-RADS grades 0 or 3 in the initial ultrasound were suggested for an MAM rescreen. The results of the MAM were also classified using the BI-RADS system.
Histopathological Examination (Biopsy): Participants who received BI-RADS grades 4 or 5 in either the initial ultrasound or MAM were advised to undergo further biopsy examination. For those with MAM rescreen results of grade 0 or grade 3, a short-term follow-up (3–6 months) or further biopsy examination was recommended based on the evaluation of breast specialists.
Follow-Up: Participants who received BI-RADS levels 1 or 2 in either the ultrasound or MAM were monitored closely through visits and phone calls conducted by trained medical social workers.
Fig. 3.
Breast Cancer Screening Flowchart Based on “Two Cancers (Breast and Cervical Cancer) Screening” Project. “Initial ultrasound, MG rescreen” was applied
The final results of the screening tests (positive or negative) were determined using a two-step approach. Initial results were categorized as negative, indeterminate, or positive based on BI-RADS grades assessed by ultrasound. Participants with indeterminate initial results were reclassified to negative or positive based on follow-up MAM assessments. Initial screening results were determined by the BI-RADS grade from the breast ultrasound: results were considered negative with BI-RADS grades 1 or 2, positive with grades 4 or 5, and indeterminate with grades 0 or 3. Participants with indeterminate results were informed of the detection of suspicious lesions that required further MAM assessment. The results from this second step were classified as negative or positive based on the MAM BI-RADS grade.
After receiving a negative final screening result, participants did not undergo any additional diagnostic procedures. However, those with a positive final screening result were referred for further diagnostic work-up to exclude or confirm a diagnosis of breast cancer at FUSCC.
Interval cancers were defined as cases diagnosed following a negative or indeterminate screening test, with no subsequent mammography or diagnostic workup within 1 year, confirmed by linkage to regional cancer registries. To assess interval cancers, we collected data on all breast cancers diagnosed, along with an additional year of follow-up from the Shanghai Municipal Cancer Registry, one of the largest cancer registries globally and an associate member of the International Association of Cancer Registries (IARC). For each patient diagnosed with breast cancer outside of the screening (i.e., diagnosed with interval cancer), we gathered all relevant medical records. Two experienced radiologists reviewed both the screening ultrasound and MAM images from the study and the clinical MAM and ultrasound images used for breast cancer diagnosis, reaching a consensus on whether the breast cancer could be retrospectively identified in the screening.
Outcomes
The primary endpoint of this study was improved screening sensitivity, enabling the detection of more true-positive cases. True-positive screening results were defined as positive screening tests in participants with histologically confirmed breast cancer through subsequent diagnostic evaluation. Screening sensitivity was calculated as the proportion of screen-detected breast cancers among all breast cancer cases (comprising both screen-detected and interval cancers) identified within one year of follow-up in the screened population. Screen-detected cancers were defined as breast cancer cases confirmed through pathological diagnosis following a positive screening test result. Interval cancers were defined as breast cancers diagnosed either: (1) after a negative screening test, or (2) after an indeterminate screening result without subsequent diagnostic follow-up (including mammography or other diagnostic examinations). Interval cases then included false negatives, true interval cancers, as well as cases where a woman with a prior negative screening exam presented with symptoms between a normal screening interval and was found to have cancer.
The secondary outcome was the detection of more proportion of early-stage cancers. Early-stage cancers were defined as those meeting any of the following criteria at the time of diagnosis: a size of less than 20 mm; no metastatic lymph nodes in the axilla; no distant metastasis; inclusion of non-invasive cancers; or classified as stage 0, stage I, or stage II according to the American Joint Committee on Cancer (AJCC) 8th Edition staging system [29].
Sample size consideration
Routine ultrasound screening for breast cancer has a sensitivity of 40%. We hypothesize that AI-assisted ultrasound screening will achieve a sensitivity of at least 70%. Assuming a significance level (α) of 0.05, 80% power, and a 5‰ breast lesion detection rate among Asian women, the estimated total sample size required is 16,800 participants—with 8400 allocated to the conventional screening group and 8400 to the AI-assisted screening group.
Statistical analysis
Numerical data are presented as medians with Interquartile Range (IQR), while categorical variables are expressed as percentages. Sensitivity was calculated by dividing the number of true-positive screenings by the total number of true positives and false positives. Specificity was determined by dividing the number of true-negative screenings by the total number of true negatives and false negatives. The PPV was estimated by dividing the number of participants with true-positive screenings by the total number of participants with positive screenings. Conversely, the NPV was calculated by dividing the number of participants with true-negative screenings by the total number of participants with negative screenings. To calculate 95% confidence intervals (CIs), we employed bootstrapping based on 5,000 samples. The significance of differences in incidence rates was assessed using Poisson regression. For categorical variables, we utilized Fisher’s exact test or the likelihood-based χ2 test. All analyses were conducted using IBM SPSS (version 20).
Results
Screening participants
Between January 1, 2021 and December 31, 2022, a total of 21,790 individuals were included in this study. The median age of the participants was 62 years (IQR, 55–66). At cluster selection, Hongkou was assigned to receive AI-assisted ultrasound screening (AI group, n = 8736), while Pudong was assigned to receive routine ultrasound screening (routine group, n = 13,054). The median age in the AI group was 64 years (IQR, 60–67), compared to 59 years (IQR, 50–65) in the routine group (Table 1). The AI group was statistically older than the routine group (P < 0.001).
Table 1.
Characteristics of screening participants
| Characteristics | AI-assisted ultrasound screening | Routine ultrasound screening | All |
|---|---|---|---|
| N | 8736 | 13,054 | 21,790 |
| Age | |||
| Median (IQR) | 64 (60–67) | 59 (50–65) | 62(55–66) |
| Age Distribution — no./total no. (%) | |||
| < 45 yr | 96(1.1) | 1845(14.1) | 1941(8.9) |
| 45–49 yr | 54(0.6) | 1141(8.7) | 1195(5.5) |
| 50–54 yr | 433(5.0) | 1869(14.3) | 2302(10.6) |
| 55–59 yr | 1362(15.6) | 2212(16.9) | 3574(16.4) |
| 60–64 yr | 3071(35.2) | 2708(20.7) | 5779(26.5) |
| ≥ 65 yr | 3720(42.6) | 3279(25.1) | 6999(32.1) |
Screening results
Among 21,790 screened participants, a total of 232 tested positive, resulting in a screening positivity rate of 10.7‰, of which the AI-assisted screening identified 107 positives, yielding a positivity rate of 12.2‰, whereas the routine screening identified 125 positives, with a positivity rate of 9.6‰. No significant difference in screen-positive cases was observed between the two groups (P = 0.07). Ultimately, the screen-positive results led to a diagnosis of breast cancer in 30 participants (true positives), corresponding to a cancer detection rate of 1.4‰ (see Table 2). Additionally, 19 participants (0.9‰) were diagnosed with interval breast cancer.
Table 2.
Screening test performance
| AI-assisted ultrasound screening | Routine ultrasound screening | All | |
|---|---|---|---|
| Screened participants | 8736 | 13,054 | 21,790 |
| Screen-Positive test result | 107 | 125 | 232 |
| True positive | 21 | 9 | 30 |
| False positive | 86 | 116 | 202 |
| Screen-Negative test result | 8629 | 12,929 | 21,558 |
| True negative | 8622 | 12,917 | 21,539 |
| False negative | 7 | 12 | 19 |
| Participants with screen-detected breast cancer | 21 | 9 | 30 |
| Detected cancers per 1000 screened | 2.4 | 0.7 | 1.4 |
| Number of interval cancers | 7 | 12 | 19 |
| Interval cancers per 1000 screened | 0.8 | 0.9 | 0.9 |
| Ratio of interval to detected | 0.3:1 | 1.3:1 | 0.63:1 |
| Sensitivity(95% CI) | 75.0(54.8,88.6) | 42.8(22.6,65.6) | 61.2(46.2,74.5) |
| Specificity(95% CI) | 99.0(98.8,99.2) | 99.1(98.9,99.3) | 99.1(98.9,99.2) |
| Positive predictive value(95% CI) | 19.6(12.8,28.7) | 7.2(3.6,13.6) | 12.9(9.0,18.1) |
| Negative predictive value(95% CI) | 99.9(99.8,99.9) | 99.9(99.8,99.9) | 99.9(99.8,99.9) |
Table 2 provides the screening characteristics for each group. The AI group demonstrated a significantly higher screening sensitivity (75%,95% CI 54.8–88.6; 21 true positives) compared to the routine group (42.8%, 95% CI 22.6–65.6; 9 true positives).
Breast cancer
After a one-year follow-up, a total of 49 cases of breast cancer were diagnosed among all screening participants. Of these, 61.2% (30 out of 49) of breast cancers were detected on screening, while 38.8% (19 out of 49) were interval cancers. The AI screening demonstrated a higher cancer detection rate compared to the routine screening group (AI: 2.4‰, 21 of 8,736; routine: 0.7‰, 9 of 13,054; P = 0.001), while no significant difference was found between the two groups regarding interval cancer detection (AI: 0.8‰,7 of 8736; routine: 0.9‰, 12 of 13,054; P = 0.789). (Table 2).
Further age-stratified analysis showed that the AI group achieved significantly better breast cancer detection rate specifically in women aged 60 and above (Table 3).
Table 3.
Screening test performance, by age group
| AI-assisted ultrasound screening | Routine ultrasound screening | All | P | |
|---|---|---|---|---|
| < 50 yr | ||||
| Screened participants | 150 | 2986 | 3136 | |
| Screen-Positive (‰) | 2(13.3) | 20(6.7) | 22(7.0) | 0.369 |
| Screen-detected breast cancer(‰) | 0(0.0) | 3(1.0) | 3(1.0) | 0.863 |
| Interval cancer(‰) | 0(0.0) | 1(0.3) | 1(0.3) | 0.952 |
| 50–59 yr | ||||
| Screened participants | 1795 | 4081 | 5876 | |
| Screen-Positive (‰) | 15(8.3) | 34(8.3) | 49(8.3) | 0.976 |
| Screen-detected breast cancer(‰) | 3(1.7) | 3(0.7) | 6(1.0) | 0.341 |
| Interval cancer(‰) | 3(1.7) | 3(0.7) | 6(1.0) | 0.341 |
| 60- yr | ||||
| Screened participants | 6791 | 5987 | 12,778 | |
| Screen-Positive (‰) | 90(13.3) | 71(11.9) | 161(12.6) | 0.486 |
| Screen-detected breast cancer(‰) | 18(2.7) | 3(0.5) | 21(1.6) | 0.002 |
| Interval cancer(‰) | 4(0.6) | 8(1.3) | 12(0.9) | 0.186 |
The clinical characteristics of the detected breast cancers for each screening group were presented in Table 4. Screen-detected breast cancers were significantly more likely to be diagnosed at stages 0, I, or II (93.3%), whereas only 68.8% of participants with interval breast cancers were diagnosed at these stages. Furthermore, the proportion of early-stage cancers among screen-detected cases was significantly higher in the AI group (95.2%, 20/21) than in the routine group (88.9%, 8/9; P < 0.001). Additionally, stage IV cancer was diagnosed in 5.3% of participants with interval breast cancers, while none of the screen-detected breast cancers were diagnosed at stage IV.
Table 4.
Clinical characteristics of screen-detected and interval cancers, by disease stage
| AI-assisted ultrasound screening | Routine ultrasound screening | All | P value | |
|---|---|---|---|---|
| Detected cancer | 21 | 9 | 30 | |
| Positive rate ‰ | 2.4 | 0.7 | 1.4 | 0.001 |
| Dcis | 3(14.3) | 0 | 3(10.0) | |
| I | 11(52.4) | 5(55.6) | 16(53.3) | |
| II | 6(28.6) | 3(33.3) | 9(30.0) | |
| III | 1(4.7) | 1(11.1) | 2(6.7) | |
| IV | 0 | 0 | 0 | |
| Early breast camcer (Dcis + I + II) * | 20(95.2) | 8(88.9) | 28(93.3) | < 0.001 |
| Interval cancer | 7 | 12 | 19 | |
| Positive rate ‰ | 0.8 | 0.9 | 0.9 | 0.789 |
| Dcis | 1(14.2) | 0 | 1(5.3) | |
| I | 2(28.6) | 4(33.3) | 6(31.6) | |
| II | 2(28.6) | 6(50.1) | 8(42.0) | |
| III | 2(28.6) | 1(8.3) | 3(15.8) | |
| IV | 0 | 1(8.3) | 1(5.3) | |
| Early interval breast camcer (Dcis + I + II) | 5(57.2) | 10(76.7) | 15(68.8) | 0.616 |
| All breast cancer | 28 | 21 | 49 | |
| Positive rate ‰ | 3.2 | 1.6 | 2.2 | 0.017 |
| Dcis | 4(14.3) | 0 | 4(8.2) | |
| I | 13(46.4) | 9(42.9) | 22(44.9) | |
| II | 8(28.6) | 9(42.9) | 17(34.7) | |
| III | 3(10.7) | 2(9.5) | 5(10.2) | |
| IV | 0 | 1(4.7) | 1(2.0) | |
| All early breast camcer (Dcis + I + II) | 25(89.3) | 18(85.7) | 43(87.8) | 0.018 |
*The proportion of early-stage breast cancer is defined as the number of early-stage cases (including DCIS, Stage I and Stage II) divided by the total number of screen-detected breast cancer cases
Discussion
In this study, we explored the characteristics of ultrasound-based breast cancer screening in China, comparing AI-assisted real-time ultrasound with routine ultrasound in a large-scale population-based screening initiative. Out of 21,790 participants, 30 (1.4‰) were diagnosed with breast cancer through screening, and another 19 (< 1‰) were identified with interval breast cancer. The AI-assisted approach detected significantly more breast cancers than the routine screening (AI: 21 of 8736; routine: 9 of 13,054, P = 0.001). Screen-detected breast cancers were substantially more often diagnosed at early stages (including stages 0, I, or II), and the proportion of early-stage cancers among screen-detected cases was significantly higher in the AI group than in the routine group.
Breast cancer screening is widely implemented to detect early abnormalities and effectively reduce mortality rates. Current screening guidelines in Europe and the United States recommend mammography, while the use of ultrasound remains contentious, particularly for Asian women with dense breast tissue. Research indicates that screening ultrasound can lead to a 30% absolute increase in the detection of invasive cancers in women with dense breasts [30]. The ACRIN 6666 trial (ClinicalTrials.gov NCT00072501) demonstrated that the breast cancer detection rate via ultrasound was comparable to that of mammography, revealing a higher proportion of invasive and node-negative cancers. This suggested that screening ultrasound could serve as a viable alternative to mammography, particularly in regions lacking organized screening programs and where low-cost, portable ultrasound systems are available [31]. A multi-center randomized trial in China (ClinicalTrials.gov NCT01880853) [17] comparing ultrasound with mammography among high-risk Chinese women found that ultrasound screening had higher sensitivity (100% for ultrasound vs. 57.1% for mammography, P = 0.04) and diagnostic accuracy (0.999 vs. 0.766, P = 0.01). Notably, ultrasound screening (also known as "two-cancer screening") has made great achievements in rural areas of China [27, 28].
In this study, we identified 30 breast cancers among 21,790 participants, resulting in a cancer detection rate of 1.37 per 1000, which is significantly lower than detection rates reported in Western countries, including the United States (7.85 per 1000 in the National Breast and Cervical Cancer Early Detection Program, NBCCEDP 2022) [32], the United Kingdom (8.69 per 1000 in the National Health Service Breast Screening Program, NHSBSP 2021–2022) [33], Canada (2.5–7.9 per 1000 in the Canadian National Breast Screening Study) [34, 35], as well as Japan (3.2–5.0 per 1000) [36]. However, our detection rate was higher than that of the Chinese National Breast Cancer Screening Program (CNBCSP) from 2009 to 2011 (0.6 per 1000 for urban women and 0.5 per 1000 for rural women) [27] and in 2015 (0.85 per 1000 for rural women) [28]. Several reasons could lead to the lower detection rates, including varying baseline breast cancer incidence across populations and differing screening strategies, particularly the screening techniques used. For instance, mammography screening achieved higher detection rates in Western countries, and Japan also reported higher detection rates for mammography alone (3.2 per 1000) and in combination with ultrasound (5.0 per 1000) [36]. In China, either initial screenings conducted via clinical breast examination (CBE) followed by MAM or ultrasound from 2009, or initially screened by ultrasound and followed by MAM post-2012, maintained low detection rates [27, 28]. Thus, there is an urgent need to explore new strategies to increase detection rates in China.
Routine ultrasound examinations may lead to misdiagnosis due to variability in operator skills and subjective judgment. In contrast, AI systems, trained on extensive datasets using deep learning, could identify subtle tumor changes, thereby improving breast cancer lesion detection. Our findings indicate that AI-assisted ultrasound screening significantly increased the cancer detection rate, as evidenced by the AI screening group detecting more cases than the routine screening group (AI: 21 of 8736; routine: 9 of 13,054, P = 0.001). The sensitivity of the AI screening group was 75.0% (95% CI 54.8–88.6), higher than that of the routine screening group (42.8%, 95% CI 22.6–65.6), as well as J-START (30.2%, 95% CI 23.9–36.5) [36] and ACRIN 6666 (52.3%, 95% CI 43.2–61.3) [31]. Furthermore, the specificity with AI was 99.0% (95% CI 98.8–99.2), comparable to the routine screening specificity (99.1%, 95% CI 98.9–99.3), yet significantly higher than ACRIN 6666 (86.3%, 95% CI 86.1–87.8) [33]. Notably, the predictive value of positive screening test results in the AI screening group was 19.6% (95% CI 12.8–28.7). Although it implies more than 80% of the participants were referred for false-positive results, it still surpassed the routine screening group (7.2%, 95% CI 3.6–13.6) and ACRIN 6666 (4.3%, 95% CI 2.9–6.4) [31]. The integration of AI significantly enhanced the accuracy and sensitivity of ultrasound image analysis, particularly in the early detection of breast cancer. Both sensitivity and specificity observed in the current study are promising for future use in large-scale population.
Despite the advantages of AI-assisted ultrasound in breast cancer screening, we noted a higher incidence of interval cancers (AI: 7 of 8736; routine: 12 of 13,054, P = 0.789). This is primarily due to the limited application of ultrasound as the main screening tool among Chinese women, which may also be related to the specific characteristics of breast cancer incidence and breast composition among Chinese women. It underscores the need for close monitoring and follow-up of cases within screening intervals, suggesting that increasing screening frequency could optimize screening effectiveness.
There are several limitations in this study. Firstly, our study was a prospective clustered and controlled study. The clustered design and units’ differences may introduce additional biases and confounding factors, such as age-related confounding and socioeconomic/residential factors, potentially affecting the comparison as well as limiting the generalizability of the results. Therefore, we conducted age-stratified analyses (detailed in Table 3) and Shanghai’s rapid urban development has minimized socioeconomic/residential disparities among districts. We believed differences in cancer incidence between two districts are insignificant. Secondly, breast cancer detection rates vary by population risk level. Since our study failed to collect risk factors and prior screening history information, our results may be affected by variations in breast cancer risks among participants. However, as our study population was from a general community sample, we consider the included participants to represent an average-risk community population. Thirdly, the project was based on current "two-cancer screening," resulting in a participant pool predominantly comprising housewives and fewer employed individuals, which may introduce selection bias regarding population representativeness. And information about follow-up MAMs was mainly gathered through telephone surveys, potentially leading to information bias. Fourthly, the interpretation of stage distribution differences should be cautious due to the limited number of advanced cancers detected, though the proportional trend aligns with AI's higher sensitivity for small lesions. Lastly, the 1-year follow-up period was relatively short, possibly underestimating breast cancer incidence and hindering evaluations of differences in screening characteristics between 1- and 2-year follow-ups, thus leading to impossibility to assess of the relationship between screening frequency and effectiveness.
Nonetheless, this study has several strengths. It is a large, population-based, prospective screening trial that enhances the generalizability and applicability of the findings. Moreover, AI-assisted ultrasound screening exhibited superior sensitivity and specificity characteristics for breast cancer detection, demonstrating excellent population identification capabilities and highlighting AI's potential in early breast cancer detection. Future researches may focus on the long-term impacts of AI technology in breast cancer screening, and improving screening accuracy to maximize early detections and improve survival rates.
Conclusions
In conclusion, AI technology presents significant advantages in this context, and broader implementation of AI-assisted ultrasound in large-scale population-based breast cancer screening is anticipated to enhance screening effectiveness.
Acknowledgements
The authors wish to thank registrars from the Shanghai Cancer Registry and staff members of the research teams for their contributions and support for the study.
Abbreviations
- AI
Artificial intelligence
- WHO
World health organization
- MAM
Mammography
- FUSCC
Fudan university shanghai cancer center
- BI-RADSB
Breast imaging reporting and data system
- IARC
International association of cancer registries
- AJCC
American joint committee on cancer
- PPV
Positive predictive value
- NPV
Negative predictive value
- IQR
Interquartile range
- CI
Confidence interval
- NBCCEDP
National breast and cervical cancer early detection program
- NHSBSP
National health service breast screening program
- CNBCSP
Chinese national breast cancer screening program
- CBE
Clinical breast examination
Author contributions
Conception and design: Y.Z. and C.C.; Development of methodology: J.S., Y.L., A.L., and X.G.; Acquisition of data: J.Z., C. W., P.J., L.Z., and C.Y.; Analysis and interpretation of data: M.M, X. F., and Z. W.; Writing, review, and/or revision of the manuscript: all authors; Study supervision: Z. X., W.Y., S.Z., Y.Z., and C.C.
Funding
Shanghai Aging and Maternal and Child Health Research Project (No: 2020YJZX0206).
Data availability
The data that support the findings of this study are available on request from the corresponding author (Y.Z.). The data are not publicly available due to privacy and ethical restrictions.
Declarations
Ethical approval
The study was performed according to the guidelines of the Helsinki Declaration and was approved by the Medical Ethics Committee of Fudan University Shanghai Cancer Center (No. 2008223-22). Informed consent was obtained from all individual participants included in the study.
Competing interests
The authors declare no potential conflicts of interest.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Jie Shen, Yajing Liu, Aihong Liu and Xiaoqin Gu have contributed equally to this work.
Contributor Information
Zhenyu Xie, Email: xiezhenyu@sina.com.
Wen Yao, Email: yw612423@163.com.
Shichong Zhou, Email: sczhou@hotmail.com.
Ying Zheng, Email: zhengying@fudan.edu.cn.
References
- 1.Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209–49. [DOI] [PubMed] [Google Scholar]
- 2.Han BF, Zheng RS, Zeng HM, Wang SM, Sun KX, Chen R, et al. Cancer incidence and mortality in China, 2022. J Natl Cancer Center. 2024;4(1):47–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Moss SM, Wale C, Smith R, Evans A, Cuckle H, Duffy SW. Effect of mammographic screening from age 40 years on breast cancer mortality in the UK Age trial at 17 years’ follow-up: a randomised controlled trial. Lancet Oncol. 2015;16:1123–32. [DOI] [PubMed] [Google Scholar]
- 4.Habbema JD, van Oortmarssen GJ, van Putten DJ, Lubbe JT, van der Maas PJ. Age-specific reduction in breast cancer mortality by screening: an analysis of the results of the Health Insurance plan of greater New York study. J Natl Cancer Inst. 1986;77(2):317–20. [PubMed] [Google Scholar]
- 5.Nelson HD, Fu R, Cantor A, Pappas M, Daeges M, Humphrey L. Effectiveness of breast cancer screening: systematic review and meta-analysis to update the 2009 US preventive services task force recommendation. Ann Intern Med. 2016;164(4):244–55. [DOI] [PubMed] [Google Scholar]
- 6.Nyström L, Andersson I, Bjurstam N, Frisell J, Nordenskjöld B, Rutqvist LE. Long-term effects of mammography screening: updated overview of the Swedish randomised trials. Lancet. 2002;359(9310):909–19. [DOI] [PubMed] [Google Scholar]
- 7.Massat NJ, Dibden A, Parmar D, Cuzick J, Sasieni PD, Duffy SW. Impact of screening on breast cancer mortality: the UK program 20 years on. Cancer Epidemiol Biomarkers Prev. 2016;25(3):455–62. [DOI] [PubMed] [Google Scholar]
- 8.Plevritis SK, Munoz D, Kurian AW, Stout NK, Alagoz O, Near AM, et al. Association of screening and treatment with breast cancer mortality by molecular subtype in US women, 2000–2012. JAMA. 2018;319(2):154–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Bleyer A, Welch HG. Effect of three decades of screening mammography on breast-cancer incidence. N Engl J Med. 2012;367(21):1998–2005. [DOI] [PubMed] [Google Scholar]
- 10.Siegel RL, Miller KD, Jemal A. Cancer statistics, 2020. CA Cancer J Clin. 2020;70(1):7–30. [DOI] [PubMed] [Google Scholar]
- 11.World health Organization WHO position paper on mammography screening. https://www.who.int/publications/i/item/who-position-paper-on-mammography-screening accessed 2024-07-11]
- 12.Yan H, Ren W, Jia M, Xue P, Li Z, Zhang S, et al. Breast cancer risk factors and mammographic density among 12518 average-risk women in rural China. BMC Cancer. 2023;23(1):952. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ontario Health (Quality). Supplemental screening as an adjunct to mammography for breast cancer screening in people with dense breasts: a health technology assessment. Ont Health Technol Assess Ser. 2023 23(9):1–293. [PMC free article] [PubMed]
- 14.Wang Y, Chen H, Li N, Ren J, Zhang K, Dai M, et al. Ultrasound for breast cancer screening in high-risk women: results from a population-based cancer screening program in China. Front Oncol. 2019;24(9):286. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Zhang X, Lin X, Tan Y, Zhu Y, Wang H, Feng R, et al. A multicenter hospital-based diagnosis study of automated breast ultrasound system in detecting breast cancer among Chinese women. Chin J Cancer Res. 2018;30(2):231–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Nothacker M, Duda V, Hahn M, Warm M, Degenhardt F, Madjar H, et al. Early detection of breast cancer: benefits and risks of supplemental breast ultrasound in asymptomatic women with mammographically dense breast tissue. A systematic review. BMC Cancer. 2009;9:335. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Shen S, Zhou Y, Xu Y, Zhang B, Duan X, Huang R, et al. A multi-centre randomised trial comparing ultrasound vs mammography for screening breast cancer in high-risk Chinese women. Br J Cancer. 2015;112(6):998–1004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Breast cancer Professional Committee of Chinese Anti-Cancer Association. Screening and early diagnosis of breast cancer in China: a practice guideline. China Onc. 2022;32(4):363–72. [Google Scholar]
- 19.Wang P, Berzin TM, Glissen Brown JR, Bharadwaj S, Becq A, Xiao X, et al. Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study. Gut. 2019. 10.1136/gutjnl-2018-317500. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Ardila D, Kiraly AP, Bharadwaj S, Choi B, Reicher JJ, Peng L, et al. End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med. 2019;25(8):1319. [DOI] [PubMed] [Google Scholar]
- 21.McKinney SM, Sieniek M, Godbole V, Godwin J, Antropova N, Ashrafian H, et al. International evaluation of an AI system for breast cancer screening. Nature. 2020;577:89–94. [DOI] [PubMed] [Google Scholar]
- 22.Lehman CD, Wellman RD, Buist DS, Kerlikowske K, Tosteson AN, Miglioretti DL, et al. Diagnostic accuracy of digital screening mammography with and without computer-aided detection. JAMA Intern Med. 2015;175:1828–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Fei XY, Shen L, Ying SH, Cai YH, Zhang Q, Kong WT, et al. Parameter transfer deep neural network for single-modal B-mode ultrasound-based computer-aided diagnosis. Cogn Comput. 2020;12:1252–64. [Google Scholar]
- 24.Qiao MY, Fang Z, Guo Y, Zhou SC, Chang C, Wang YY. Breast calcification detection based on multichanneradio frequency signals via a unified deep learning framework. Expert Syst Appl. 2021;168:114218. [Google Scholar]
- 25.Guo Y, Hu Y, Qiao M, Wang Y, Yu J, Li J, et al. Radiomics analysis on ultrasound for prediction of biologic behavior in breast invasive ductal carcinoma. Clin Breast Cancer. 2018;18(3):e335-344. [DOI] [PubMed] [Google Scholar]
- 26.Shen J, Liu YJ, Mo Mo, Zhou J, Wang ZZ, Chang C. Effectiveness of artificial intelligence-assisted ultrasound for breast cancer screening in Chinese women (in Chinese). China onc. 2023;33(11):1002–8. [Google Scholar]
- 27.Ma L, Lian ZQ, Zhao YX, Di JL, Song B, Ren WH, et al. Breast ultrasound optimization process analysis based on breast cancer screening for 1501753 rural women in China. Chin J Oncol. 2021;43(4):497–503. [DOI] [PubMed] [Google Scholar]
- 28.Huang Y, Dai H, Song F, Li H, Yan Y, Yang Z, et al. Preliminary effectiveness of breast cancer screening among 1.22 million Chinese females and different cancer patterns between urban and rural women. Sci Rep. 2016;6:39459. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.American Joint Committee on Cancer. Breast. In: AJCC Cancer Staging Manual. 8th ed. New York, NY: Springer; 2017:589.
- 30.Warner E. Clinical practice breast-cancer screening. N Engl J Med. 2011;365(11):1025–32. [DOI] [PubMed] [Google Scholar]
- 31.Berg WA, Bandos AI, Mendelson EB, Lehrer D, Jong RA, Pisano ED. Ultrasound as the primary screening test for breast cancer: analysis from ACRIN 6666. J Natl Cancer Inst. 2015;108(4):3djv67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.National Breast and Cervical Cancer Early Detection Program (NBCCEDP): https://www.cdc.gov/breast-cervical-cancer-screening/about/index.html
- 33.Breast Screening Programme, NHS England, National statistics. https://digital.nhs.uk/data-and-information/publications/statistical/breast-screening-programme/england---2022-23
- 34.Miller AB, Baines CJ, To T, Wall C. Canadian national breast screening study: 1. Breast cancer detection and death rates among women aged 40–49 years. CMAJ. 1992;147(10):1459–76. [PMC free article] [PubMed] [Google Scholar]
- 35.Miller AB, Baines CJ, To T, Wall C. Canadian national breast screening study: 2. Breast cancer detection and death rates among women aged 50–59 years. CMAJ. 1992;147(10):1477–88. [PMC free article] [PubMed] [Google Scholar]
- 36.Ohuchi N, Suzuki A, Sobue T, Kawai M, Yamamoto S, Zheng YF, et al. J-START investigator groups sensitivity and specificity of mammography and adjunctive ultrasonography to screen for breast cancer in the Japan strategic anti-cancer randomized trial (J-START): a randomised controlled trial. Lancet. 2016;387(10016):341–8. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author (Y.Z.). The data are not publicly available due to privacy and ethical restrictions.