Artificial intelligence-amplified contrast enhancement in brain magnetic resonance imaging for improving image quality and lesion visualization: a prospective pilot study.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, The Affiliated Hospital of Southwest Medical University, Luzhou, China.
- Department of Radiology, Sichuan Provincial People's Hospital, University of Electronic Science and Technology of China, Chengdu, China.
- Institute of Radiation Medicine, Sichuan Provincial People's Hospital, University of Electronic Science and Technology of China, Chengdu, China.
Abstract
Artificial intelligence (AI) algorithms synthesizing virtual standard-dose images from low-dose contrast-enhanced images of brain magnetic resonance imaging (MRI) have been repurposed to boost contrast from standard-dose input. This study aimed to prospectively evaluate the impact of a Food and Drug Administration (FDA)-cleared, deep learning-based software on contrast enhancement, lesion visualization, and diagnostic confidence for standard-dose contrast-enhanced images. This prospective study enrolled patients undergoing contrast-enhanced brain MRI between August 2025 and September 2025. Precontrast and standard-dose postcontrast three-dimensional T1-weighted (T1w) images were acquired. AI-amplified contrast-enhanced images were generated via an FDA-cleared deep learning software based on precontrast and standard postcontrast images. Two radiologists independently performed quantitative analyses, examining contrast-to-noise ratio (CNR), lesion-to-brain ratio (LBR), and contrast enhancement percentage (CEP). Qualitative assessments of lesion border delineation, internal morphology, and contrast enhancement, and diagnostic confidence were performed with a 4-point Likert scale. Comparisons between AI-amplified and standard-dose images were conducted via the Wilcoxon signed-rank test. Forty-one patients (mean age 51.1±12.6 years) with enhancing brain lesions were included. For both readers, AI-amplified images, as compared with standard contrast-enhanced images, exhibited a significantly higher CNR (82.56±41.15 <i>vs</i>. 20.33±23.28; 87.91±50.30 <i>vs</i>. 20.76±12.00), LBR (2.86±0.64 <i>vs</i>. 1.78±0.48; 2.83±0.53 <i>vs</i>. 1.76±0.40), and CEP (Reader 1: 393.00%±188.40% <i>vs</i>. 201.20%±111.32%; Reader 2: 382.26%±176.71% <i>vs</i>. 197.57%±109.90%) (all P values <0.001). Compared with standard-dose images, AI-amplified images produced an approximately 400% higher CNR, a 60% higher LBR, and a 200% higher CEP. The qualitative evaluation of both readers indicated that the AI-amplified images provided significantly improved lesion border delineation (Reader 1: 3.38±0.61 <i>vs</i>. 3.13±0.59; Reader 2: 3.46±0.60 <i>vs</i>. 3.00±0.63) and contrast enhancement (Reader 1: 3.88±0.31 <i>vs</i>. 2.94±0.24; Reader 2: 3.85±0.42 <i>vs</i>. 2.90±0.49) (all P values <0.001). No significant differences were observed in lesion internal morphology scores between AI-amplified and standard-dose images (P>0.05 for both readers). AI-amplified images demonstrated significantly higher diagnostic confidence than did standard-dose images for both readers (Reader 1: 3.59±0.50 <i>vs</i>. 3.20±0.60; Reader 2: 3.48±0.64 <i>vs</i>. 3.05±0.74; both P values <0.001). AI-based contrast amplification significantly increases the quantitative contrast metrics, qualitative lesion visualization, and diagnostic confidence for standard-dose contrast-enhanced brain MRI without increasing the gadolinium dose. These findings support the use of AI-based contrast amplification as a complementary tool in routine clinical neuroimaging.