Related Experiment Video
Updated: Jan 9, 2026
Concurrent EEG and Functional MRI Recording and Integration Analysis for Dynamic Cortical Activity Imaging
Published on: June 30, 2018
PISCO: Self-supervised k-space regularization for improved neural implicit k-space representations of dynamic MRI.
Veronika Spieker1, Hannah Eichhorn2, Wenqi Huang3
1Institute of Machine Learning in Biomedical Imaging, Helmholtz Munich, Munich, Germany; School of Computation, Information and Technology, Technical University of Munich (TUM), Munich, Germany; Millenium Institute for Intelligent Healthcare Engineering, Santiago, Chile.
We introduce PISCO, a self-supervised k-space loss function, to improve neural implicit k-space representations (NIK) for dynamic MRI. PISCO enhances reconstruction quality, especially at high acceleration factors, by enforcing k-space consistency without extra data.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Signal Processing
Background:
- Neural implicit k-space representations (NIK) show promise for dynamic MRI.
- Reduced acquisition time leads to overfitting and performance degradation in NIK.
Purpose of the Study:
- To introduce a novel self-supervised k-space loss function, LPISCO, for NIK regularization.
- To improve the performance of NIK-based dynamic MRI reconstructions, particularly under accelerated acquisition conditions.
Main Methods:
- Developed a self-supervised k-space loss function, parallel imaging-inspired self-consistency (PISCO).
- Enforced consistent global k-space neighborhood relationships within NIK.
- Evaluated PISCO on static and dynamic MR reconstructions.
Main Results:
- PISCO significantly improved NIK representations and dynamic MRI reconstruction quality.
- Superior spatio-temporal reconstruction was achieved at high acceleration factors (R ≥ 50).
- NIK with PISCO avoided temporal oversmoothing compared to state-of-the-art methods.
Conclusions:
- PISCO is a versatile self-supervised k-space loss function for NIK.
- It enhances dynamic MRI reconstruction without compromising temporal resolution.
- PISCO demonstrates potential for broader applications in MRI reconstruction and other architectures.
