COMPARING THE EFFICACY OF AI-ASSISTED VS. TRADITIONAL DIAGNOSTIC IMAGING IN RADIOLOGY – A META-ANALYSIS

Background: Artificial intelligence (AI) is transforming diagnostic imaging in radiology by improving accuracy, efficiency, and decision-making. Traditional radiological interpretation, despite its clinical significance, is limited by interobserver variability, workload constraints, and potential diagnostic errors. While AI-assisted imaging has demonstrated superior performance in certain studies, inconsistencies in outcomes and a lack of consensus necessitate a comprehensive meta-analysis to evaluate its efficacy compared to conventional radiology. Objective: This meta-analysis aims to assess the diagnostic accuracy, sensitivity, and specificity of AI-assisted imaging compared to traditional radiology across multiple imaging modalities, including computed tomography (CT), magnetic resonance imaging (MRI), and mammography. Methods: A systematic search was conducted in PubMed, Embase, Cochrane Library, Web of Science, and Scopus for studies published between 2019 and 2024. Randomized controlled trials (RCTs), cohort studies, and systematic reviews comparing AI-assisted imaging with traditional radiological interpretation were included. A random-effects model was applied to account for study heterogeneity, and statistical measures such as standardized mean difference (SMD) and 95% confidence intervals (CIs) were used to estimate pooled effect sizes. Heterogeneity was assessed using the I² statistic, and publication bias was evaluated through funnel plot analysis and Egger’s test. Results: A total of 32 studies with a combined sample size of 130,000+ patients were included. AI-assisted imaging exhibited significantly higher diagnostic accuracy compared to conventional radiology, with pooled effect sizes ranging from SMD = 1.0 to 1.5 (p < 0.05). The highest performance was noted in AI-based detection of gastrointestinal lesions and cancer metastases. Heterogeneity was moderate to high (I² = 56.3%, p = 0.02), necessitating subgroup analyses. Funnel plot analysis suggested mild publication bias. Conclusion: AI-assisted diagnostic imaging demonstrates superior accuracy and efficiency compared to traditional radiology, supporting its integration into clinical workflows. However, variability in algorithm performance and potential biases warrant further prospective validation studies. Standardized AI implementation guidelines and human-AI collaboration strategies are necessary for optimizing its clinical utility.