A densely interconnected network for deep learning accelerated MRI

• DOI: 10.1007/s10334-022-01041-3
• Publication Date: 2022-09-14T09:11:07Z
• Authors: Jon André Ottesen; Matthan W. A. Caan; Inge Rasmus Groote; Atle Bjørnerud
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Acceleration; Image and Video Processing (eess.IV); Computer Science - Computer Vision and Pattern Recognition; 610; 02 engineering and technology; Electrical Engineering and Systems Science - Image and Video Processing; Signal-To-Noise Ratio; Magnetic Resonance Imaging; 620; 03 medical and health sciences; Deep Learning; 0302 clinical medicine; 0202 electrical engineering, electronic engineering, information engineering; FOS: Electrical engineering, electronic engineering, information engineering; Research Article

To improve accelerated MRI reconstruction through a densely connected cascading deep learning framework.A framework (reference model) was modified by applying three architectural modifications: input-level dense connections between cascade inputs and outputs, an improved sub-network, long-range skip-connections subsequent networks. An ablation study performed, where five model configurations were trained on the NYU fastMRI neuro dataset with end-to-end scheme conjunct four- eightfold acceleration. The models evaluated comparing their respective structural similarity index measure (SSIM), normalized mean square error (NMSE), peak signal to noise ratio (PSNR).The proposed interconnected residual network (DIRCN), utilizing all suggested modifications achieved SSIM improvement of 8% 11%, NMSE 14% 23%, PSNR 2% 3% for acceleration, respectively. In study, individual contributed this both acceleration factors, improving SSIM, NMSE, approximately 2-4%, 4-9%, 0.5-1%, respectively.The allow simple adjustments already existing further resulting reconstructions.