Artificial intelligence versus elastography in characterizing BI-RADS 4 breast nodules: A systematic review and critical appraisal

Breast imaging reporting and data system (BI-RADS) category 4 breast lesions represent a heterogeneous category with moderate suspicion of malignancy, which pose significant diagnostic challenges. Both artificial intelligence (AI) and elastography have demonstrated potential adjunctive roles in improving the evaluation of these lesions. Given the increasingly pervasive use of AI in the medical field, a systematic and critical evaluation of its diagnostic efficacy, clinical utility, and practical applications, compared with elastography techniques, is warranted for the assessment of BI-RADS 4 breast nodules. A systematic literature search was conducted across multiple databases from January 2010 to December 2024, and the studies were critically appraised using standardized quality assessment tools (e.g., quality assessment of diagnostic accuracy studies-2). Due to the significant heterogeneity in study populations and methodologies, a narrative synthesis approach with comprehensive critical appraisal was employed. A total of 23 studies met the inclusion criteria for AI assessment (n = 15,847 lesions) and 31 for elastography (n = 12,456 lesions). Critical appraisal revealed significant methodological limitations—67% of the studies had a high risk of bias in patient selection, 45% in index-test conduct, and 56% in flow and timing. Only 25% of the studies were considered high quality. AI systems demonstrated promising diagnostic performance in individual studies (reported area under the curve range: 0.82–0.94), while elastography showed consistent but more modest performance (reported area under the curve range: 0.72–0.87). However, the quality of the evidence was insufficient for a reliable comparative assessment. While both technologies show promise, the existing evidence is limited by significant methodological constraints, precluding reliable comparative conclusions. These gaps highlight the need for high-quality prospective head-to-head comparison studies with standardized protocols and rigorous methodology.