Congbo Cai

Objective. Quantitative magnetic resonance imaging (qMRI) can noninvasively acquire parametric maps, providing quantitative biomarkers for clinical assessment. Parametric maps, such as T ₂ , T ₂ ^* , and apparent diffusion coefficient (ADC), characterize distinct biophysical processes in tissue in an interpretable, numerical manner. Owing to their distinct physical origins, various parametric maps differ in physical units, numerical magnitudes, and dynamic ranges. Even for the same type of parametric map, inter-subject differences may still exist in anatomical morphology, tissue characteristics, and the local distributions of parameter values. However, deep learning methods trained in image-domain directly learn pixel-level parameter values, and variations in numerical scales and subject-specific parameter distributions may result in anatomical detail loss and localized artifacts in certain tasks. To address this challenge and to deliver dependable quantitative references for clinical assessment, we propose a wavelet-guided diffusion method. Approach. We extend the discrete convolution wavelet transform (DCWT) to two dimensions and combine it with a conditional denoising diffusion method to develop a diffusion-based DCWT (Diff-DCWT) for parameter mapping. Diff-DCWT achieves accurate parameter mapping under the guidance of wavelet coefficient maps, thereby reducing reliance on image domain learning. The performance of Diff-DCWT is evaluated through simulation and in vivo experiments across T ₂ , T ₂ ^* , and ADC mapping tasks. Main Results. Both simulation and healthy volunteer experiments demonstrate that T ₂ , T ₂ ^* , and ADC maps obtained from Diff-DCWT maintain excellent visual quality and quantitative accuracy. Clinical case studies suggest the potential utility of Diff-DCWT-derived T ₂ and ADC maps for stroke staging and help distinguish lesional from normal brain regions in glioma. Significance. Diff-DCWT establishes a high-fidelity framework for parameter mapping and exhibits preliminary clinical value in stroke staging and glioma characterization. Moreover, this work demonstrates the potential of wavelet-domain prior guidance for deep learning-based qMRI applications.

Objective. Magnetic susceptibility source separation provides profound insights into tissue pathophysiology by disentangling the underlying paramagnetic and diamagnetic sources within the human brain. However, the highly time-consuming nature of existing acquisition techniques creates a critical need for rapid and reliable separation methodologies to facilitate routine clinical applications. Approach. A highly accelerated framework for whole-brain magnetic susceptibility source separation was proposed by utilizing the ultra-fast 3D-MOLED sequence. This pipeline integrates the advanced deep learning architecture to accurately map the separated susceptibility components without relying on spin echo data. To ensure quantitative stability and structural fidelity, a robust three-dimensional domain adaptation strategy was incorporated during the simulation phase. The performance of the proposed method was systematically evaluated through phantom experiments, healthy volunteer assessments, and clinical examinations of patients with multiple sclerosis. Main results. The proposed framework achieved whole-brain susceptibility source separation at 1 mm isotropic resolution with only 50 s of data acquisition. In healthy volunteers, the paramagnetic and diamagnetic susceptibility maps were highly comparable to those obtained with the conventional multi-echo gradient-echo sequence, with structural similarity indices of 0.9244 ± 0.0181 and 0.9425 ± 0.0186, peak signal-to-noise ratios of 33.76 ± 2.37 and 37.29 ± 3.51 dB, regional coefficients of determination of 0.973 and 0.934, and Bland–Altman biases of −0.0013 and −0.0010 ppm, respectively. Furthermore, clinical evaluations in patients with multiple sclerosis demonstrated consistent characterization of complex lesion-related susceptibility features. Significance. Achieving whole-brain magnetic susceptibility source separation in less than one minute represents a meaningful technical advance. By substantially reducing the acquisition time required for susceptibility source separation, this accelerated framework may improve the feasibility of future clinical translation in neuroimaging workflows.

Objective. To address the challenges of rapid and robust quantitative MRI, particularly for T _{1 ρ} mapping, by developing and evaluating a novel technique—spin-locked (SL) multiple overlapping echo detachment (SL-MOLED)—for efficient mapping of T _{1 ρ} , T ₂ , and T ₂ ^* relaxation times with reduced sensitivity to B ₀ / B ₁ inhomogeneities and SL –related banding artifacts. Approach. SL-MOLED integrates SL preparation into the MOLED acquisition framework, enabling simultaneous mapping of T _{1 ρ} , T ₂ , T ₂ ^* , proton density, and estimation of Δ B ₀ and B ₁ in approximately 11 s per slice. A synthetic data-driven deep learning reconstruction framework was trained on Bloch-simulated datasets with explicitly modeled banding artifacts, allowing effective mitigation of artifact-related errors. Validation comprised numerical experiments, phantom studies, healthy volunteer experiments, and a preliminary patient evaluation. Reconstruction accuracy was assessed using the structural similarity index measure (SSIM), mean absolute error (MAE), Pearson’s correlation coefficient ( r ), and Bland–Altman analysis. Short-term within-session stability and 7 d inter-session test–retest repeatability were evaluated separately using ROI-based coefficients of variations. Main results. Numerical experiments showed that networks trained with artifact modeling improved SSIM by 0.1–0.2 and reduced MAE by 2–5 ms for T _{1 ρ} , T ₂ , and T ₂ ^* compared with models trained without artifact modeling, across varying B ₀ / B ₁ inhomogeneities and SL frequencies. Phantom studies demonstrated good agreement between SL-MOLED reconstructed maps and reference methods for T _{1 ρ} , T ₂ , and T ₂ ^* ( r > 0.997, Bland–Altman bias < 3.4%). In vivo experiments confirmed strong correlations with reference maps ( r ⩾ 0.976) and good repeatability. In a patient with multiple sclerosis, SL-MOLED detected elevated T _{1 ρ} values in lesions relative to normal-appearing white matter, while T ₂ and T ₂ ^* showed smaller changes, indicating that each parameter may reflect different underlying pathological features. Significance. SL-MOLED provides accurate, repeatable, and artifact-robust quantitative mapping within a substantially reduced acquisition time (∼11 s per slice), offering a promising framework for reliable multi-parametric MRI with potential for clinical translation.

Multi-parametric quantitative magnetic resonance imaging (mqMRI) holds significant clinical potential through multi-parametric tissue characterization, yet its adoption is hindered by prolonged scan time and sensitivity to non-ideal signal conditions, especially in high-resolution whole-brain protocols. To address these challenges, we propose a novel signal encoding method integrating phase-modulated three dimensional steady-state free precession with multiple overlapping-echo detachment (3D SSFP-MOLED). This method simultaneously encodes six physiological parameters (M₀, T₁, T₂, T₂*, B₁⁺, ΔB₀) into k-space by controlling overlapping echo detachment in signal acquisition. A physics-constrained synthetic data pipeline was developed to simulate MR signal evolutions with realistic field variations (ΔB₀, B₁⁺ inhomogeneities), enabling robust training of network for real-time parameter mapping. Whole-brain parametric maps (1×1×2 mm³ resolution) can be delivered within 3 minutes with only 2x parallel acquisition acceleration. Validation was performed on phantom, healthy volunteers, and clinical cases with tumors/hemorrhage. Experimental results show that our method can achieve rapid multi-parametric quantitation with high accuracy and reproducibility. By synergizing adaptive signal encoding, physics-informed synthetic training, and reproducible deep learning reconstruction, this work establishes a new paradigm for efficient and reliable mqMRI in clinical signal processing applications.

Background/Objectives: Accurate brain tumor segmentation from MRI is crucial for diagnosis but faces challenges like domain shifts across medical centers, data privacy constraints, and high annotation costs. While source-free active domain adaptation (SFADA) emerges as a promising solution to these issues, existing approaches often overlook the inherent structural complexity in tumor regions. Methods: We propose a novel SFADA framework composed of two major contributions. First, we introduce a Region-level Uncertainty-Guided Sample Selection (RUGS) strategy, enabling the identification of the most informative target-domain samples in a single inference pass. Second, we present the Source-Free Active Domain Adaptation Network (SFADA-Net), a Mamba-driven segmentation model equipped with a dual-path multi-kernel convolution module for enhanced local feature interaction and a structure-aware prompted Mamba module for capturing global spatial relationships. Results: Extensive evaluations across one source domain dataset (BraTS-2021) and three target domain datasets (BraTS-SSA, BraTS-PED, and BraTS-MEN 2023) demonstrate the superior adaptability of the proposed method, achieving consistently high segmentation accuracy across domains. With only 5% annotation budget, our framework consistently outperforms state-of-the-art segmentation and domain adaptation methods, achieving robust segmentation accuracy across diverse domains and approaching the performance of fully supervised learning. Conclusions: The proposed method achieves superior accuracy in brain tumor region segmentation and precise boundary delineation under a limited annotation budget. It effectively mitigates domain shift while fully complying with data privacy regulations. Consequently, our framework relieves manual annotation bottlenecks and accelerates the cross-center deployment of accurate diagnostic tools, facilitating the clinical application of domain adaptation.

Objective: Quantitative magnetic resonance imaging (qMRI) has attracted more and more attention in clinical diagnosis and medical sciences due to its capability to non-invasively characterize tissue properties. Nevertheless, most qMRI methods are time-consuming and sensitive to motion, making them inadequate for quantifying organs with physiological movement. In this context, single-shot multiple overlapping-echo detachment (MOLED) imaging technique has been presented, but its acquisition efficiency and image quality are limited when the field of view (FOV) is smaller than the object, especially for abdominal organs and myocardium. Methods: A novel single-shot reduced FOV qMRI method was developed based on MOLED (termed rFOV-MOLED). This method combines zonal oblique multislice (ZOOM) and outer volume suppression (OVS) techniques to reduce the FOV and suppress signals outside the FOV. A deep neural network was trained using synthetic data generated from Bloch simulations to achieve high-quality T₂ map reconstruction from rFOV-MOLED images. Numerical simulation, water phantom and in vivo abdominal and myocardial imaging experiments were performed to evaluate the method. The coefficient of variation and repeatability index were used to evaluate the reproducibility. Multiple statistical analyses were utilized to evaluate the accuracy and significance of the method, including linear regression, Bland-Altman analysis, Wilcoxon signed-rank test, and Mann-Whitney U test, with the p-value significance level of 0.05. Results: Experimental results show that rFOV-MOLED achieved excellent performance in reducing aliasing signals due to FOV reduction. It provided T₂ maps closely resembling the reference maps. Moreover, it gave finer tissue details than MOLED and was quite repeatable. Conclusion and Significance: rFOV-MOLED can ultrafast and stably provide accurate T₂ maps for myocardium and specific abdominal organs with improved acquisition efficiency and image quality.

Objective: Multi-parametric quantitative mag- netic resonance imaging (mqMRI) provides comprehensive and accurate information about tissue microstructure and holds significant clinical value for the diagnosis and treatment of diseases. However, conventional methods require long scan time, leading to registration errors and physiological variability between different sequence acqui- sitions. This study aims to propose an advanced imaging method that addresses these limitations. Methods: A novel approach called longitudinal magnetization controlled multiple overlapping-echo detachment (LMC-MOLED) imaging was proposed. LMC-MOLED leverages a deep neural network trained on synthetic data generated from Bloch simulation, incorporating non-ideal factors such as B₀ and B₁ inhomogeneities to efficiently reconstruct parametric maps. Results: LMC-MOLED enables simulta- neous quantification of T₁, T₂, T₂^*, proton density (PD), ΔB₀, and B₁ parameters in approximately 1.2 seconds per slice. Validation experiments using numerical brain, phantom, and human brains demonstrate its excellent performance, particularly in terms of acquisition speed, image quality, and robustness. Additionally, LMC-MOLED effectively corrects distortions introduced by long echo train acquisition. Conclusion and Significance: LMC-MOLED offers a rapid, robust solution for mqMRI, providing multi-parametric mapping in a single scan with signify- cantly reduced acquisition time. It holds potential to improve diagnostic accuracy and alleviate patient burden.