31 papers
This paper proposes Multi-Kernel Gated Adapters (MKGA) and their residual variant (ResMKGA) to address asymmetric feature degradation under cross-center domain shifts in thyroid ultrasound by leveraging the complementary strengths of CNNs and ViTs to enhance both segmentation robustness and malignancy diagnostic accuracy.
This paper presents the design, construction, and initial performance evaluation of the IMAS system, a long axial field-of-view PET prototype that uniquely combines time-of-flight and depth-of-interaction capabilities, demonstrating sub-4 mm spatial resolution and improved tumor identification in pilot clinical results despite some performance limitations attributed to data transfer bottlenecks.
This study demonstrates the feasibility of using an interleaved 7T multinuclear MRS protocol to simultaneously track brain glucose levels and high-energy phosphate metabolism, revealing that both metrics significantly increase in response to experimentally induced hyperglycemia in healthy adults.
This paper introduces a quantum mechanics-inspired interaction-picture framework to quantify the abscopal effect as a continuous, stochastic phenomenon in the context of PULSAR, enabling individual-level analysis of tumor interactions and standardized cross-study comparisons in preclinical models.
This paper proposes and optimizes a magnetically driven elastic microswimmer that achieves autonomous, nonreciprocal propulsion through hysteretic collapse of its segments, enabling the simultaneous independent control of multiple microrobots via a single oscillating magnetic field for potential medical applications.
This paper reports on the development of a high-speed, high-resolution X-ray camera based on the INTPIX4NA SOIPIX detector and a SiTCP-XG 10GbE readout system, demonstrating its successful application in three recent experiments at KEK facilities: X-ray zooming microscopy, phase-contrast imaging, and nondestructive lithium detection in battery materials.
This paper proposes "Efficient Vision Mamba," a computationally lightweight deep learning framework that combines multi-head selective state-space models with hybrid scanning to achieve state-of-the-art MRI super-resolution performance while drastically reducing parameters and computation compared to existing methods.
This paper, authored by the ISMRM MRS Synthetic Data Working Group, reviews and evaluates current practices, applications, and considerations for generating and utilizing synthetic data in Magnetic Resonance Spectroscopy to address challenges such as limited training data, software validation, and the optimization of acquisition and deep learning models.
This paper presents an extension of the GPU-accelerated Monte Carlo toolkit gPET to support flexible multi-layer detector geometries for Depth-of-Interaction (DOI) encoding, demonstrating through validation that the new framework enables efficient simulation and optimization of PET systems with significantly improved radial spatial resolution while maintaining comparable sensitivity and runtime performance.
This study demonstrates that delivering low-energy (2.5 MV) photon beams at an extended source-to-patient distance (4 m) in an upright configuration achieves significantly sharper penumbra, lower surface dose, and higher peak-to-valley ratios in spatially fractionated plans compared to standard 6 MV coplanar radiotherapy, offering a viable alternative to FLASH for high-flux sources.
GOUHFI 2.0 is an updated deep-learning toolbox that enables robust, contrast-agnostic whole-brain segmentation, cortical parcellation, and volumetry for Ultra-High Field MRI by utilizing a domain-randomization strategy and two specialized 3D U-Net networks trained on diverse datasets.
This paper presents an image recovery algorithm that utilizes multiple harmonics and total variation regularization to estimate true phase stepping positions, effectively eliminating Moiré artifacts in attenuation, differential-phase, and dark-field images generated by grating interferometers.
This paper proposes a super-resolution iterative reconstruction method for analyzer-less X-ray grating interferometry that enables high-quality multi-modal imaging with reduced dose and relaxed detector sampling requirements, overcoming the limitations of traditional algorithms.
This paper reports the first successful in-air irradiation of biological samples using stable, kHz laser-driven electron beams, demonstrating promising pre-clinical results of normal tissue sparing in zebrafish embryos while maintaining anticancer efficacy in glioblastoma cells, thereby marking a significant milestone toward the clinical translation of laser-plasma accelerators.
This paper demonstrates that the optical extinction manifold of dielectric materials exhibits intrinsic sparsity best captured by the Discrete Cosine Transform rather than the FFT, enabling a compressed sensing architecture that achieves high-fidelity material reconstruction with a 51–94% reduction in hardware sensors by overcoming traditional Nyquist limits.
This paper presents a low-cost, wireless wearable Electrical Impedance Tomography system utilizing five synchronized parallel AD5933 chips and specialized signal conditioning to achieve high-precision, real-time imaging of biological tissues such as human lungs with a signal-to-noise ratio exceeding 50 dB.
This paper introduces a novel early warning signal called DE-AC, which estimates the dominant eigenvalue from autocorrelation functions to reliably predict the onset of period-doubling bifurcations in cardiac systems with superior sensitivity and specificity compared to existing indicators.
This paper introduces a rotation-invariant graph neural network that generalizes across arbitrary diffusion MRI acquisition protocols without retraining, enabling rapid and accurate brain microstructure estimation for clinical deployment.
This feasibility study demonstrates that Sparse Probabilistic Evaluation (SPE), a computationally efficient method using a predefined error grid, achieves sufficient accuracy for probabilistic treatment planning in IMPT head and neck patients while maintaining clinically acceptable computation times.
This paper proposes a supervised conditional flow matching model that effectively synthesizes high-quality, artifact-reduced CT-like images from cone-beam CT scans, significantly improving Hounsfield unit accuracy and image metrics to enable more reliable organ segmentation and dose calculation in image-guided radiotherapy.