Artificial intelligence in radiology workflow: A systematic review into protocol automation and clinical applications

Objective To evaluate the performance, clinical integration, and workflow impact of artificial intelligence (AI) models developed for automated MRI and CT protocol selection in radiology. Methods A systematic search was conducted in PubMed, Embase, Scopus, and Medline for English-language studies published between January 2000 and February 2025. Eligible studies included original research that applied AI techniques, such as natural language processing (NLP), machine learning (ML), or deep learning (DL), to imaging protocol assignment. Two reviewers independently screened studies using Rayyan, assessed methodological quality using the QUADAS-2 tool, and extracted performance metrics, integration status, and workflow outcomes. Due to methodological heterogeneity, results were synthesised narratively. Results Fifteen studies met inclusion criteria out of 742 screened articles. Most focused on MRI protocoling, using transformer-based NLP models (e.g., BERT, GPT-3.5) with high accuracy (84%–98%, limited to MRI-specific tasks). CT studies showed similarly strong but more variable performance, ranging from ~85% in AI-based multiclass classification to ~99% concordance in rules-based institutional systems. ML models (e.g., random forest, support vector machines) also performed well with reported accuracy >83%. Protocoling time was reduced by up to 72%, and 60%–70% of outpatient referrals were handled by full or partial automation. Eight studies described real-world deployment, including integration with electronic health record systems. Common limitations included limited external validation, inconsistent reporting, and a lack of transparency regarding model decision logic. Conclusion AI-based tools for MRI and CT protocol automation demonstrate high predictive accuracy and substantial workflow benefits in radiology practice. However, broader implementation requires cross-institutional validation, standardised reporting practices, and alignment with clinical governance frameworks to ensure safe and scalable integration.