Deep learning reconstruction for chest MRI: image quality and diagnostic agreement with CT.
Authors
Affiliations (5)
Affiliations (5)
- Department of Diagnostic Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, No. 17 Panjiayuan Nanli, Chaoyang District, Beijing 100021, China.
- Department of Thoracic Surgery, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, No. 17 Panjiayuan Nanli, Chaoyang District, Beijing 100021, China.
- GE Healthcare China, No. 1 Tongji South Road, Daxing District, Beijing 100176, China.
- Department of Thoracic Surgery, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, No. 17 Panjiayuan Nanli, Chaoyang District, Beijing 100021, China. Electronic address: [email protected].
- Department of Diagnostic Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, No. 17 Panjiayuan Nanli, Chaoyang District, Beijing 100021, China. Electronic address: [email protected].
Abstract
To evaluate whether deep learning reconstruction (DLR) improves chest magnetic resonance imaging (MRI) image quality and to assess the agreement between DLR MRI and computed tomography (CT) for characterization of known pulmonary lesions. This prospective study included 54 patients with CT-visible pulmonary lesions who underwent 3.0-T chest MRI between September 2022 and September 2024. T2-weighted imaging with fat suppression and unenhanced/enhanced three-dimensional liver acquisition with volume acceleration were reconstructed with and without DLR. Quantitative image quality was assessed using background noise, signal-to-noise ratio, and contrast-to-noise ratio. Qualitative image quality was evaluated by two senior radiologists in consensus using 5-point scales for anatomical clarity, artifact severity, and lesion edge sharpness. Morphological features and lesion diameter measurements were compared between DLR MRI and CT. DLR significantly reduced background noise and improved signal-to-noise and contrast-to-noise ratios across all MRI sequences. For example, median lesion signal-to-noise ratio on T2-weighted images increased from 186.8 to 394.8 after DLR reconstruction (P < 0.001). Qualitative scores also improved significantly, with lesion edge sharpness increasing from 3.78 ± 0.46 to 4.70 ± 0.50 on T2-weighted images (P < 0.001). DLR MRI showed high agreement with CT for lobulation, nodule type, shape, and pleural indentation (κ = 0.902-1.00). Lesion size measurements showed excellent agreement between DLR MRI and CT (ICC = 0.972). DLR improves chest MRI image quality and shows high agreement with CT for assessing known pulmonary lesions, supporting its potential use as a non-ionizing adjunct for follow-up imaging.