Comparison of the adaptive unrolled method's effectiveness against two other contemporary deep unrolling techniques (M1 and M2). M1 produces a blurry reconstruction, M2 results in a clear but structurally inaccurate output. In contrast, the adaptive unrolling method achieves reconstruction that closely matches the Ground Truth in both sharpness and structural accuracy.​
Invention Summary:
Magnetic Resonance Imaging (MRI) is a widely used imaging modality for clinical diagnostics and the planning of surgical interventions. Accelerated MRI seeks to mitigate the inherent limitation of long scanning time by reducing the amount of raw k-space data required for image reconstruction. Recently, the deep unrolled model (DUM) has demonstrated significant effectiveness and improved interpretability for MRI reconstruction, by truncating and unrolling the conventional iterative reconstruction algorithms with deep neural networks. However, the potential of DUM for MRI reconstruction has not been fully exploited.
Rutgers researchers have identified improvements to key components of the DUM-based MRI reconstruction that improve the MRI reconstruction quality, speed and amount of memory required:
- Improving the adaptive gradient algorithm to achieve self-adaptive dynamic learning rates by adjusting for different spatial areas in an MR image providing a more flexible updating strategy in each iteration stage.
- Incorporating the momentum technique used in gradient descent acceleration and performing multi-stage, multi-level feature aggregation scheme to accelerate the iteration convergence.
- Use of adjacent information to improve multi-coil MRI reconstruction for accurate and memory-efficient sensitivity map estimation.
Market Applications:
- Medical imaging (MRI, Ultrasound, CT)
- AI-based diagnostics
- Image Reconstruction processes with sparse data
Advantages:
- Demonstrated effectiveness in MRI reconstruction benchmarks for knee, brain and heart.
- Potential use in Signal Processing and Telecommunications (audio signals, geophysics, seismic imaging) applications.
- Drastically improves processing speed
Publication: https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/09565.pdf
Intellectual Property & Development Status: Provisional application filed. Patent pending. Available for licensing and/or research collaboration. For any business development and other collaborative partnerships, contact: marketingbd@research.rutgers.edu
