Abstract:Early detection of many life-threatening diseases (e.g., prostate and breast cancer) within at-risk population can improve clinical outcomes and reduce cost of care. While numerous disease-specific "screening" tests that are closer to Point-of-Care (POC) are in use for this task, their low specificity results in unnecessary biopsies, leading to avoidable patient trauma and wasteful healthcare spending. On the other hand, despite the high accuracy of Magnetic Resonance (MR) imaging in disease diagnosis, it is not used as a POC disease identification tool because of poor accessibility. The root cause of poor accessibility of MR stems from the requirement to reconstruct high-fidelity images, as it necessitates a lengthy and complex process of acquiring large quantities of high-quality k-space measurements. In this study we explore the feasibility of an ML-augmented MR pipeline that directly infers the disease sidestepping the image reconstruction process. We hypothesise that the disease classification task can be solved using a very small tailored subset of k-space data, compared to image reconstruction. Towards that end, we propose a method that performs two tasks: 1) identifies a subset of the k-space that maximizes disease identification accuracy, and 2) infers the disease directly using the identified k-space subset, bypassing the image reconstruction step. We validate our hypothesis by measuring the performance of the proposed system across multiple diseases and anatomies. We show that comparable performance to image-based classifiers, trained on images reconstructed with full k-space data, can be achieved using small quantities of data: 8% of the data for detecting multiple abnormalities in prostate and brain scans, and 5% of the data for knee abnormalities. To better understand the proposed approach and instigate future research, we provide an extensive analysis and release code.

Adaptive Sampling of k-Space in Magnetic Resonance for Rapid Pathology Prediction

The MRI Scanner as a Diagnostic: Image-less Active Sampling

On the Feasibility of Machine Learning Augmented Magnetic Resonance for Point-of-Care Identification of Disease

PILOT: Physics-Informed Learned Optimized Trajectories for Accelerated MRI

Deep End-to-end Adaptive k-Space Sampling, Reconstruction, and Registration for Dynamic MRI

Statistically Segregated k-Space Sampling for Accelerating Multiple-Acquisition MRI

End-to-End Sequential Sampling and Reconstruction for MR Imaging.

Undersampled Dynamic MRI Reconstruction by Double Sparse Spatiotemporal Dictionary.

Patient-Adaptive and Learned MRI Data Undersampling Using Neighborhood Clustering

Adaptive Convolutional Neural Networks for Accelerating Magnetic Resonance Imaging Via K -Space Data Interpolation

End-to-End Sequential Sampling and Reconstruction for MRI

Learn Stable MRI Under-Sampling Pattern With Decoupled Sampling Preference

LSST: Learned Single-Shot Trajectory and Reconstruction Network for MR Imaging

Bayesian Reconstruction of Magnetic Resonance Images using Gaussian Processes

Learning Optimal K-space Acquisition and Reconstruction Using Physics-Informed Neural Networks

Development of Fast Magnetic Resonance Imaging Techniques Based on k-space Accelerated Collection

Towards Ultrafast MRI via Extreme k-Space Undersampling and Superresolution

Towards performant and reliable undersampled MR reconstruction via diffusion model sampling

Classification, Regression and Segmentation directly from k-Space in Cardiac MRI

Adaptive Local Neighborhood-based Neural Networks for MR Image Reconstruction from Undersampled Data

Reconstruction Of Undersampled Big Dynamic Mri Data Using Non-Convex Low-Rank And Sparsity Constraints