AI-accelerated compressed sensing and deep learning reconstruction for colon MRI: prospective validation of image quality and preliminary assessment of staging accuracy versus contrast-enhanced CT.
Authors
Affiliations (4)
Affiliations (4)
- Department of Radiology, Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Peking University Cancer Hospital and Institute, No. 52 Fu Cheng Road, Hai Dian District, 100142, Beijing, China.
- Department of Radiology, Peking University Shougang Hospital, No.9 Jin Yuan Zhuang Road, Shi Jing Shan District, 100144, Beijing, China.
- Shanghai United Imaging Healthcare Co., Ltd, No. 2258, Chengbei Road, Jiading District, 201807, Shanghai, China.
- Department of Radiology, Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Peking University Cancer Hospital and Institute, No. 52 Fu Cheng Road, Hai Dian District, 100142, Beijing, China. [email protected].
Abstract
To prospectively validate an AI-accelerated MRI protocol combining L₁-regularized compressed sensing and deep learning reconstruction for colon cancer staging, compared with contrast-enhanced CT (CECT). This single-center study (August 2022-April 2024) enrolled 132 patients with pathologically confirmed colon cancer who underwent both 3.0-T non-contrast MRI and CECT within 1 week. Patients unable to complete an MRI were excluded. MRI sequences included T1- and T2-weighted imaging and diffusion-weighted imaging. Image quality was assessed using a 4-point Likert scale, alongside quantitative analysis of contrast-to-noise ratio (CNR) and edge rise distance (ERD). Surgical pathology correlation was available for 33 cases. Five radiologists independently reviewed the images. Among 153 matched examination pairs from 132 patients (mean age, 58 ± 11 years), AI-MRI outperformed CECT in both qualitative and quantitative measures. MRI yielded higher image quality scores (4 [IQR 3-4] vs. 3 [IQR 3-3]), improved CNR (5.5 [IQR 4.2-7.0] vs. 1.3 [IQR 0.7-2.2]), and reduced ERD (1.6 mm [IQR 1.2-2.8] vs. 3.7 mm [IQR 2.8-5.0]). MRI provided superior peritoneal sharpness (75.8%) and colonic wall visualization (100%), and superior or equivalent tumor conspicuity (92.2%) and nodal assessment (86.9%), whereas CECT demonstrated better artifact control (45.1%). In preliminary analyses, MRI achieved higher T-staging accuracy (78.8% vs. 30.3%) and N-staging specificity (93.3% vs. 13.3%). The AI-accelerated MRI protocol significantly outperformed CECT in image quality and preliminary staging accuracy for colon cancer evaluation while eliminating radiation exposure. Question Can an AI-accelerated MRI protocol provide superior image quality and staging performance compared with contrast-enhanced CT, which has limited diagnostic accuracy for colon cancer staging? Findings AI-accelerated MRI using L₁-regularized compressed sensing and deep learning reconstruction achieved superior image quality and higher preliminary staging accuracy than contrast-enhanced CT. Critical Relevance Statement AI-accelerated MRI improves image quality and staging confidence without radiation exposure, supporting its potential to enhance preoperative assessment and clinical decision-making in colon cancer.