EEG-based connectomes offer a low-cost and portable method to identify brain disorders using deep learning. With the growing interest in model interpretability and transparency, explainable artificial intelligence (XAI) is widely applied to understand the decision of deep learning models. However, most research focuses solely on interpretability analysis based on the insights from XAI, overlooking XAI’s potential to improve model performance. To bridge this gap, we propose a dynamical-system-inspired architecture, XAI guided transformer (XAIguiFormer), where XAI not only provides explanations but also contributes to enhancing the transformer by refining the originally coarse information in self-attention mechanism to capture more relevant dependency relationships. In order not to damage the connectome’s topological structure, the connectome tokenizer treats the single-band graphs as atomic tokens to generate a sequence in the frequency domain. To address the limitations of conventional positional encoding in understanding the frequency and mitigating the individual differences, we integrate frequency and demographic information into tokens via a rotation matrix, resulting in a richly informative representation. Our experiment demonstrates that XAIguiFormer achieves superior performance over all baseline models. In addition, XAIguiFormer provides valuable interpretability through visualization of the frequency band importance. Our code is available at https://github.com/HanningGuo/XAIguiFormer.

MR measurement using a combination of X-nuclei and proton MRI is of great interest as the information provided by the two nuclei is highly complementary, with the X-nuclei signal giving metabolic data relating to potential biomarkers and the proton signal affording anatomical details. Due to the relatively weak signal obtained from X-nuclei, combining an X-nuclei coil with a proton coil is also advantageous for [Formula: see text] shimming and scout images. One approach to building a double-resonant coil is to modify the coil geometry. Here, to achieve double-resonance, a 2× 1 ladder network was designed and tuned at both proton and X-nuclei frequencies successfully. Due to coupling between closed wires, the double-tuned coil generates a shifted transmit efficiency pattern compared to that of the single-tuned loop at the 7T MRI proton frequency. To compensate for the shifted pattern, one part of the 2× 1 ladder network was folded, and the tuning and performance of the folded double-tuned coil were evaluated in simulations and MR measurements. The proposed structure was further evaluated with overlapped decoupling in a receive-only array. The results show that our proposed folded double-tuned coil moderated the shifted pattern of a straight double-tuned loop coil and provided minimum losses at both proton and X-nuclei frequencies. The proposed folded double-tuned loop coil has also been further extended to a receive-only array.

Simultaneous MR-PET/-SPECT is an emerging technology that capitalises on the invaluable advantages of both modalities, allowing access to numerous sensitive tracers and superior soft-tissue contrast alongside versatile functional imaging capabilities. However, to optimise these capabilities, concurrent acquisitions require the MRI antenna located inside the PET/SPECT field-of-view to be operated without compromising any aspects of system performance or image quality compared to the stand-alone instrumentation. Here, we report a novel gamma-radiation-transparent antenna concept. The end-fed J-shape antenna is particularly adept for hybrid ultra-high field MR-PET/-SPECT applications as it enables all highly attenuating materials to be placed outside the imaging field-of-view. Furthermore, this unique configuration also provides advantages in stand-alone MR applications by reducing the amount of coupling between the cables and the antenna elements, and by lowering the potential specific absorption rate burden. The use of this new design was experimentally verified according to the important features for both ultra-high field MRI and the 511 keV transmission scan. The reconstructed attenuation maps evidently showed much lower attenuation ( ∼ 15 %) for the proposed array when compared to the conventional dipole antenna array since there were no high-density components. In MR, it was observed that the signal-to-noise ratio from the whole volume obtained using the proposed array was comparable to that acquired by the conventional array which was also in agreement with the simulation results. The unique feature, J-shape array, would enable simultaneous MR-PET/-SPECT experiments to be conducted without unduly compromising any aspects of system performance and image quality compared to the stand-alone instrumentation.

Simultaneous MR-PET is an increasingly popular multimodal imaging technique that is able to combine metabolic information obtained from PET with anatomical/functional information from MRI. One of the key technological challenges of the technique is the integration of a PET-transparent MR coil system, a solution to which is demonstrated here for a double-tuned1H/31P head coil at 3 T. Two single-resonant birdcage coils tuned to the1H and31P resonances were arranged in an interleaved fashion and electrically decoupled with the use of trap circuits. All high 511 keV quanta absorbing components were arranged outside the PET field-of-view in order to minimize count rate reduction. The materials inside the PET field-of-view were carefully evaluated and chosen for minimum impact on the PET image quality. As far as possible, the coil case was geometrically optimized to avoid sharp transitions in attenuation, which may potentially result in streaking artefacts during PET image reconstruction. The coil caused a count rate loss of just above 5% when inserted into the PET detector ring. Except for the anterior region, which was designed to maintain free openings for increased patient comfort, an almost uniform distribution of 511 keV attenuation was maintained around the circumference of the coil. MR-related performance for both nuclei was similar or slightly better than that of a commercial double-tuned coil, despite the MR-PET coil having a close-fitting RF screen to shield the PET and MR electronics from possible electromagnetic interferences.

The kinetic analysis of [Formula: see text]-FET time-activity curves (TAC) can provide valuable diagnostic information in glioma patients. The analysis is most often limited to the average TAC over a large tissue volume and is normally assessed by visual inspection or by evaluating the time-to-peak and linear slope during the late uptake phase. Here, we derived and validated a linearized model for TACs of [Formula: see text]-FET in dynamic PET scans. Emphasis was put on the robustness of the numerical parameters and how reliably automatic voxel-wise analysis of TAC kinetics was possible. The diagnostic performance of the extracted shape parameters for the discrimination between isocitrate dehydrogenase (IDH) wildtype (wt) and IDH-mutant (mut) glioma was assessed by receiver-operating characteristic in a group of 33 adult glioma patients. A high agreement between the adjusted model and measured TACs could be obtained and relative, estimated parameter uncertainties were small. The best differentiation between IDH-wt and IDH-mut gliomas was achieved with the linearized model fitted to the averaged TAC values from dynamic FET PET data in the time interval 4-50 min p.i.. When limiting the acquisition time to 20-40 min p.i., classification accuracy was only slightly lower (-3%) and was comparable to classification based on linear fits in this time interval. Voxel-wise fitting was possible within a computation time ≈ 1 min per image slice. Parameter uncertainties smaller than 80% for all fits with the linearized model were achieved. The agreement of best-fit parameters when comparing voxel-wise fits and fits of averaged TACs was very high (p < 0.001).

We present an evolution-strategy based approach to solve the magnitude least squares (MLS) design problem of low flip-angle slice-selective parallel transmit RF pulses for ultra-high field MRI using SAR and peak-RF-constraints. A combined transmit k-space trajectory and RF pulse weight optimization is proposed in two algorithmic steps. The first step is a coarse grid search to find an initial solution that fulfills all constraints for the subsequent multistage optimization. This avoids convergence to the next nearest local minimum. The second step attempts to refine the results using multiple evolution strategies. We compare the performance of our approach with the non-convex optimization methods described in the literature. The proposed algorithm converges for phantom and in vivo data and only requires an initial estimate of the range of suitable regularization parameters. It demonstrates improved excitation homogeneity compared to published spoke-design methods and allows optimization for homogeneity with a subsequent reduction in the SAR burden. Moreover, excitation homogeneity and the SAR burden can be balanced against each other, enabling a further reduction in SAR at the cost of minor relaxations in excitation homogeneity. This feature makes the algorithm a good candidate for SAR limited sequences in ultra-high field imaging. The algorithm is validated using phantom and in vivo measurements obtained with a 16-channel transmit array at 9.4T.

Accurate scatter correction is essential for qualitative and quantitative PET imaging. Until now, scatter correction based on Monte Carlo simulation (MCS) has been recognized as the most accurate method of scatter correction for PET. However, the major disadvantage of MCS is its long computational time, which makes it unfeasible for clinical usage. Meanwhile, single scatter simulation (SSS) is the most widely used method for scatter correction. Nevertheless, SSS has the disadvantage of limited robustness for dynamic measurements and for the measurement of large objects. In this work, a newly developed implementation of MCS using graphics processing unit (GPU) acceleration is employed, allowing full MCS-based scatter correction in clinical 3D brain PET imaging. Starting from the generation of annihilation photons to their detection in the simulated PET scanner, all relevant physical interactions and transport phenomena of the photons were simulated on GPUs. This resulted in an expected distribution of scattered events, which was subsequently used to correct the measured emission data. The accuracy of the approach was validated with simulations using GATE (Geant4 Application for Tomography Emission), and its performance was compared to SSS. The comparison of the computation time between a GPU and a single-threaded CPU showed an acceleration factor of 776 for a voxelized brain phantom study. The speedup of the MCS implemented on the GPU represents a major step toward the application of the more accurate MCS-based scatter correction for PET imaging in clinical routine.

A new parameter estimation algorithm, MERLIN, is presented for accurate and robust multiexponential relaxometry using magnetic resonance imaging, a tool that can provide valuable insight into the tissue microstructure of the brain. Multi-exponential relaxometry is used to analyze the myelin water fraction and can help to detect related diseases. However, the underlying problem is ill-conditioned, and as such, is extremely sensitive to noise and measurement imperfections, which can lead to less precise and more biased parameter estimates. MERLIN is a fully automated, multi-voxel approach that incorporates state-of-the-art ℓ1-regularization to enforce sparsity and spatial consistency of the estimated distributions. The proposed method is validated in simulations and in vivo experiments, using a multi-echo gradient-echo (MEGE) sequence at 3 T. MERLIN is compared to the conventional single-voxel ℓ2-regularized NNLS (rNNLS) and a multivoxel extension with spatial priors (rNNLS + SP), where it consistently showed lower root mean squared errors of up to 70 percent for all parameters of interest in these simulations.

Sodium (23Na) MRI provides complementary cellular and metabolic information. However, the intrinsic MR sensitivity of23Na is considerably lower compared with that of the proton, making it difficult to measure MR-detectable sodium signals. It is therefore important to maintain the signal-to-noise ratio (SNR) of the sodium signal as high as possible. Double-tuned coils are often employed in combinationwith a1H coil, providing structural images and B0 shimming capability. The double-tuned coil design can be achieved with the use of two geometrically decoupled coils whose B1 field directions are perpendicular to each other. This can be used to design quadrature-driven, single-nucleus coils to improve SNR, and additionally, this coil can also be utilized as a linear-driven double-resonant mode. Here, we have developed and evaluateda quadrature-enhanced, double-tuned coil. The novel coil uses PIN-diode switches, inserted only in the loop coil, to shift the resonance frequency between1H and23Na so that23Na signals can be acquired in quadrature and the capability of using1H function remains. Consequently, the23Na SNR values obtained with the double-tuned coil are nearly 33% and 17% higher in comparison with geometrically identical single-tuned coils. SNR plots also show the superiority of double-tuned coil in23Na.

MRI using nuclei other than protons is of clinical interest due to the important role of these nuclei in cellular processes. Phosphorous-31 (31P), for example, plays an important role in energy metabolism. However, measurement of31P can be challenging, as the receive signal is weak compared with that of proton (1H). Consequently, it is often necessary to integrate1H elements for localizations and B0shimming in RF coils intended for31P measurements. Good decoupling between the1H and the31P elements is therefore essential. In this paper, bent dipole antennas tuned to1H were integrated with a four channel31P loop coil array, in a manner providing strong geometric decoupling between dipoles and loops. As the physical length of a resonant dipole antenna is too long at 3T, the dipole antennas were bent around the load. The loss of31P elements due to the presence of the dipole antennas was evaluated by measuring scattering parameters and comparing the SNR of31P spectra with and without the presence of the dipole antennas. The performance of the bent dipole antenna was evaluated by simulation and sensitivity measurement. The Q-factors and the SNR of the four-loop array were reduced by less than 5% when the bent dipole antennas were introduced. The measured sensitivity of the bent dipole was higher (15%) than that of dual-tuned birdcage. The combined bent dipole and loop array is therefore a promising design for1H/31P applications at 3T.

A new reconstruction method, coined MIRAGE, is presented for accurate, fast, and robust parameter mapping of multiple-echo gradient-echo (MEGE) imaging, the basis sequence of novel quantitative magnetic resonance imaging techniques such as water content and susceptibility mapping. Assuming that the temporal signal can be modeled as a sum of damped complex exponentials, MIRAGE performs model-based reconstruction of undersampled data by minimizing the rank of local Hankel matrices. It further incorporates multi-channel information and spatial prior knowledge. Finally, the parameter maps are estimated using nonlinear regression. Simulations and retrospective undersampling of phantom and in vivo data affirm robustness, e.g., to strong inhomogeneity of the static magnetic field and partial volume effects. MIRAGE is compared with a state-of-the-art compressed sensing method, -ESPIRiT. Parameter maps estimated from reconstructed data using MIRAGE are shown to be accurate, with the mean absolute error reduced by up to 50% for in vivo results. The proposed method has the potential to improve the diagnostic utility of quantitative imaging techniques that rely on MEGE data.

In hybrid magnetic resonance-positron emission tomography (MR-PET) studies with the Siemens 3T MR-BrainPET scanner an instantaneous reduction of the PET sensitivity was observed during execution of certain MR sequences. This interference was investigated in detail with custom-made as well as standard clinical MR sequences. The radio-frequency pulses, the switched gradient fields and the constant magnetic field were examined as the relevant parameters of the magnetic resonance imaging (MRI) system as well as the air temperature within the PET detectors. Our investigation comprised the analysis of the analog PET signals, the total count rates, the geometric distribution of the count rate reduction within the BrainPET detector as well as reconstructed images. The fast switching magnetic field gradients were identified to distort the analog PET detector signals. The measured count rate reduction was found to be less than 3%, but only up to 2% in the case of echo planar imaging sequences, as applied in functional MRI. For clinical sequences routinely used in hybrid MR-BrainPET measurements, a correction method has been designed, implemented, and evaluated .

For iterative, fully 3D positron emission tomography (PET) image reconstruction intrinsic symmetries can be used to significantly reduce the size of the system matrix. The precalculation and beneficial memory-resident storage of all nonzero system matrix elements is possible where sufficient compression exists. Thus, reconstruction times can be minimized independently of the used projector and more elaborate weighting schemes, e.g., volume-of-intersection (VOI), are applicable. A novel organization of scanner-independent, adaptive 3D projection data is presented which can be advantageously combined with highly rotation-symmetric voxel assemblies. In this way, significant system matrix compression is achieved. Applications taking into account all physical lines-of-response (LORs) with individual VOI projectors are presented for the Siemens ECAT HR+ whole-body scanner and the Siemens BrainPET, the PET component of a novel hybrid-MR/PET imaging system. Measured and simulated data were reconstructed using the new method with ordered-subset-expectation-maximization (OSEM). Results are compared to those obtained by the sinogram-based OSEM reconstruction provided by the manufacturer. The higher computational effort due to the more accurate image space sampling provides significantly improved images in terms of resolution and noise.

A new algorithm is proposed for computing the discrete Fourier Transform (DFT) of purely phase encoded data acquired during Magnetic Resonance Imaging (MRI) experiments. These experiments use the SPRITE (Single Point Ramped Imaging with T 1 Enhancement) method and multiple-point acquisition, sampling data in a nonuniform manner that prohibits reconstruction by fast Fourier transform. The chirp z-transform algorithm of Rabiner, Schafer, and Rader can be combined with phase corrections to compute the DFT of this data to extremely high accuracy. This algorithm outperforms the interpolation methods that are traditionally used to process nonuniform data, both in terms of execution time and in terms of accuracy as compared to the DFT.