Breast cancer screening plays a central role in reducing cancer-specific mortality, yet its expanding scope has brought unintended consequences. As imaging technologies have become more sensitive, the balance between detecting malignancy and avoiding unnecessary intervention has become increasingly fragile. False-positive findings, short-interval surveillance and benign biopsies now account for a substantial share of screening-related activity. These effects are shaped less by the imaging modality itself than by interpretation thresholds and management rules, particularly those embedded in the breast imaging reporting and data system (BI-RADS). The cumulative burden of over-detection affects patients, clinicians and healthcare systems alike, with implications for psychological wellbeing, resource use and sustainability. In response, artificial intelligence has emerged as a tool capable of supporting more selective, risk-adapted screening by refining decision-making without undermining cancer detection.
Imaging Modalities and Threshold-Driven Over-Detection
Across mammography, digital breast tomosynthesis (DBT), ultrasound and magnetic resonance imaging (MRI), gains in sensitivity have frequently been accompanied by reductions in specificity. In mammography-based programmes, recall rates can approach double digits, while only a small fraction of recalled women are ultimately diagnosed with cancer. DBT reduces tissue overlap and has demonstrated modest improvements in recall rates, yet cumulative probabilities of false-positive recalls and biopsies remain high over repeated screening rounds. Differences between national and regional programmes illustrate that recall patterns depend heavily on local thresholds and reader behaviour rather than on technology alone.
Ultrasound exemplifies the challenge of threshold-driven over-detection. Its sensitivity for benign nodules often leads to BI-RADS 3 or 4A assessments, categories associated with low malignancy rates but significant follow-up activity. When pathways favour biopsy over surveillance, benign pathology dominates outcomes. MRI, while effective in reducing interval cancers among selected high-risk or extremely dense-breast populations, generates high absolute false-positive rates and considerable downstream workload when applied broadly. These patterns demonstrate that over-detection is primarily a product of how imaging findings are managed, with small shifts in recall or biopsy criteria producing large population-level effects.
Must Read: Classifying Nonmass Lesions on Breast Ultrasound
BI-RADS Management and the Consequences of Over-Surveillance
BI-RADS categories determine whether women undergo surveillance or invasive assessment. BI-RADS 3 lesions, defined by a very low likelihood of malignancy, are typically managed with short-interval follow-up. Longitudinal data show malignancy rates comparable to negative examinations, supporting extended surveillance intervals in selected cases. BI-RADS 4A lesions represent a more significant challenge. Although classified as low suspicion, they account for a large proportion of biopsy recommendations despite malignancy rates indicating that most findings are benign. This category is a major contributor to over-intervention in routine practice.
The downstream consequences of these management choices are multifaceted. False-positive results are associated with persistent anxiety and reduced quality of life, and women experiencing repeated false positives are less likely to return for future screening. Those who disengage face higher risks of interval cancers, undermining the intended benefits of screening. Economic impacts are substantial, driven by cascades of imaging, procedures and clinic visits that consume limited radiology capacity. Environmental costs further compound the issue, as avoidable imaging contributes to excess carbon emissions. Clinically, over-detection leads to overtreatment, including unnecessary biopsies and surgeries that expose patients to pain, complications and cosmetic effects despite benign outcomes.
AI-Supported Screening and Risk-Adapted Follow-Up
AI has demonstrated the capacity to address these challenges by improving consistency and precision in breast screening. In population-based programmes and prospective trials, AI-assisted reading has maintained or increased cancer detection while reducing recall rates, false positives and radiologist workload. Improvements in positive predictive value and detection of small tumours have been achieved alongside substantial reductions in double reading, indicating gains in efficiency without loss of safety.
Beyond screening reads, AI supports more granular risk stratification at the lesion level. Deep-learning models applied to mammography and ultrasound have shown the ability to safely downgrade a significant proportion of BI-RADS 4A lesions to low-risk categories, reducing unnecessary biopsies while maintaining high sensitivity. AI-driven risk forecasting extends this approach by predicting future cancer risk over several years, enabling personalised screening intervals and modality selection. These capabilities address key weaknesses of non-AI risk models, including inconsistent application and limited integration into clinical workflows.
Implementation remains complex. AI systems require local calibration, integration with existing infrastructure and ongoing monitoring to ensure performance remains stable across populations and vendors. Nevertheless, evidence from real-world deployments indicates that, when carefully governed, AI can reduce benign recalls and biopsies, ease workforce pressures and support adherence to evidence-based thresholds.
Over-detection and over-surveillance in breast screening arise from the interaction between sensitive imaging technologies and conservative management thresholds. The resulting psychological, economic, environmental and clinical burdens highlight the need for more selective, risk-adapted approaches. AI offers a practical means to rebalance sensitivity and specificity by refining BI-RADS decision-making, improving lesion stratification and enabling personalised follow-up strategies. While robust validation and thoughtful implementation are essential, integrating AI into screening pathways provides healthcare systems with an opportunity to preserve the life-saving benefits of early detection while minimising avoidable harm.
Source: Insights into Imaging
Image Credit: iStock
