Abstract
Accelerated MRI aims to reduce scan time by acquiring data more efficiently, for example, through optimized pulse sequences or readouts that increase k -space coverage per excitation (e.g., echo planar imaging), or by collecting partial k -space measurements with advanced reconstruction methods. Acceleration via partial k -space acquisition (i.e., undersampling) has received significant attention, particularly with the rise of learning-based reconstruction methods. Recent works have explored population-adaptive sampling patterns learned from groups of patients (or scans), which enhance sampling pattern design by tailoring it to dataset-specific characteristics, rather than relying on generic approaches. Building on this idea, sampling techniques can be further personalized down to the level of individual scans, enabling the capture of subject- or slice-specific details that may be overlooked in population-based designs. To address this challenging problem, we propose a framework for jointly learning scan-adaptive Cartesian undersampling patterns and a corresponding reconstruction model from a training set, enabling more tailored sampling for individual scans. We use an alternating algorithm for learning the sampling patterns and the reconstruction model where we use an iterative coordinate descent (ICD) based offline optimization of scan-adaptive k -space sampling patterns for each example in the training set. A nearest neighbor search is then used to select the scan-adaptive sampling pattern at test time from initially acquired low-frequency k -space information. We applied the proposed framework (dubbed SUNO) to the fastMRI multi-coil knee and brain datasets, demonstrating improved performance over the currently used undersampling patterns at both 4× and 8× acceleration factors in terms of both visual quality and quantitative metrics. The code for the proposed framework is available at https://github.com/sidgautam95/adaptive-sampling-mri-suno.