Enhancing Radiologist Productivity with Artificial Intelligence in Magnetic Resonance Imaging (MRI): A Narrative Review

Diagnostics, Journal Year: 2025, Volume and Issue: 15(9), P. 1146 - 1146

Published: April 30, 2025

Artificial intelligence (AI) shows promise in streamlining MRI workflows by reducing radiologists' workload and improving diagnostic accuracy. Despite MRI's extensive clinical use, systematic evaluation of AI-driven productivity gains remains limited. This review addresses that gap synthesizing evidence on how AI can shorten scanning reading times, optimize worklist triage, automate segmentation. On 15 November 2024, we searched PubMed, EMBASE, MEDLINE, Web Science, Google Scholar, Cochrane Library for English-language studies published between 2000 focusing applications MRI. Additional searches grey literature were conducted. After screening relevance full-text review, 67 met inclusion criteria. Extracted data included study design, techniques, productivity-related outcomes such as time savings The categorized into five themes: scan automating segmentation, optimizing workflow, decreasing general time-saving or reduction. Convolutional neural networks (CNNs), especially architectures like ResNet U-Net, commonly used tasks ranging from segmentation to automated reporting. A few also explored machine learning-based automation software and, more recently, large language models. Although most demonstrated efficiency accuracy, limited external validation dataset heterogeneity could reduce broader adoption. offer potential enhance radiologist productivity, mainly through accelerated scans, streamlined workflows. Further research, including prospective standardized metrics, is needed enable safe, efficient, equitable deployment tools practice.

Language: Английский

Deep Learning-Based DCE-MRI Automatic Segmentation in Predicting Lesion Nature in BI-RADS Category 4

Deleted Journal, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 25, 2024

To investigate whether automatic segmentation based on DCE-MRI with a deep learning (DL) algorithm enabled advantages over manual in differentiating BI-RADS 4 breast lesions. A total of 197 patients suspicious lesions from two medical centers were enrolled this study. Patients treated at the First Hospital Qinhuangdao between January 2018 and April 2024 included as training set (n = 138). Lanzhou University Second assigned to an external validation 59). Areas delineated DL segmentation, evaluated consistency through Dice correlation coefficient. Radiomics models constructed segmentations predict nature Meanwhile, was by both professional radiologist non-professional radiologist. Finally, area under curve value (AUC) accuracy (ACC) used determine which prediction model more effective. Sixty-four malignant cases (32.5%) 133 benign (67.5%) The DL-based showed high achieving coefficient 0.84 ± 0.11. radiomics demonstrated superior predictive performance compared radiologists, AUC 0.85 (95% CI 0.79-0.92). significantly reduced working time improved efficiency 83.2% further demonstrating its feasibility for clinical applications. outperformed radiologists distinguishing category 4, thereby helping avoid unnecessary biopsies. This groundbreaking progress suggests that is expected be widely applied practice near future, providing effective auxiliary tool diagnosis treatment cancer.

Language: Английский