Revolutionizing breast cancer screening : integrating artificial intelligence with clinical examination for targeted care in South Africa

Abstract

INTRODUCTION : Breast cancer remains a critical public health concern globally, with early detection being pivotal to improving outcomes through clinical downstaging. In low- and middle-income countries, access to traditional screening methods like mammography is limited due to high costs, infrastructure deficits, and shortages of trained professionals. This study evaluates the integration of Breast AI, an artificial intelligence (AI)-enhanced diagnostic tool, with Clinical Breast Examination (CBE) to improve breast cancer screening in resource-limited settings. Although the system demonstrated clinical utility, challenges such as cost-effectiveness, infrastructure readiness, and provider training for scaling this technology warrant further exploration. AIM AND OBJECTIVES : This study aimed to assess the clinical utility of the Breast AI system in conjunction with CBE for breast cancer screening. Objectives included evaluating the system's diagnostic performance, its potential to achieve clinical downstaging, and its ability to reduce unnecessary surgical referrals. The study also aimed to identify areas for improvement, such as logistical barriers and scaling feasibility. METHODS : A prospective comparative cohort study was conducted at Daspoort PoliClinic in Gauteng Province over 6 months. A total of 1,617 women aged 25 to 85 years were screened using CBE and Breast AI. Data collection included risk stratification, Breast Imaging Reporting and Data System (BIRADS) scoring, and referral outcomes. Statistical analyses compared the diagnostic performance of CBE and Breast AI using McNemar's test, with a Chi-square value of 1.8 and a p value of 0.1797. Educational sessions on breast cancer awareness were also conducted to encourage community engagement. RESULTS : Of the 1,617 women, 530 presented with clinical signs or risk factors. Eight patients required short-term follow-up for BIRADS-3 findings, five of whom were identified by Breast AI, compared to two identified by CBE. No cases were classified as BIRADS-5 requiring immediate intervention. The Breast AI system demonstrated improved sensitivity, identifying four additional positive cases compared to CBE, thereby reducing false negatives. Risk stratification by Breast AI ranged between 0 and 25%, indicating a low probability of malignancy but ensuring accurate referral for symptomatic cases. The system facilitated timely surgical opinions for conditions like accessory breast tissue with lipoma that CBE had missed. Despite these findings, logistical and cost-effectiveness barriers to scaling the technology remain unaddressed. CONCLUSION : The integration of Breast AI into screening programs showed promise in enhancing diagnostic accuracy, achieving clinical downstaging, and reducing unnecessary surgical referrals. The system's adjunctive use with CBE demonstrated potential for streamlining health-care delivery in resource-limited settings. However, the study highlights the need for further research on scaling this technology, addressing logistical challenges, and evaluating its cost-effectiveness. Future efforts should focus on expanding the sample population, integrating AI-driven tools into national screening protocols, and enhancing provider training to optimize patient outcomes and resource allocation.

HIGHLIGHTS • Breast cancer remains the most common form of cancer among women worldwide, with a particularly high incidence in low- and middle-income countries such as South Africa. • The study aims to develop a point-of-care risk stratified breast cancer screening program through the comparison of clinical breast examination and artificial intelligence breast ultrasound techniques in the Gauteng region. • The statistical output of the current research study holds much promise as a potential method of clinical downstaging and screening for patients located in rural settlements.