Quantifying biological parameters of interest through dynamic positron emission tomography (PET) requires an arterial input function (AIF) conventionally obtained from arterial blood samples. The AIF can also be non-invasively estimated from blood pools in PET images, often identified using co-registered MRI images. Deploying methods without blood sampling or the use of MRI generally requires total body PET systems with a long axial field-of-view (LAFOV) that includes a large cardiovascular blood pool. However, the number of such systems in clinical use is currently much smaller than that of short axial field-of-view (SAFOV) scanners. We propose a data-driven approach for AIF estimation for SAFOV PET scanners, which is non-invasive and does not require MRI or blood sampling using brain PET scans. The proposed method was validated using dynamic ¹⁸F-fluorodeoxyglucose [¹⁸F]FDG total body PET data from 10 subjects. A variational inference-based machine learning approach was employed to correct for peak activity. The prior was estimated using a probabilistic vascular MRI atlas, registered to each subject's PET image to identify cerebral arteries in the brain. The estimated AIF using brain PET images (IDIF-Brain) was compared to that obtained using data from the descending aorta of the heart (IDIF-DA). Kinetic rate constants (K₁, k₂, k₃) and net radiotracer influx (K_i) for both cases were computed and compared. Qualitatively, the shape of IDIF-Brain matched that of IDIF-DA, capturing information on both the peak and tail of the AIF. The area under the curve (AUC) of IDIF-Brain and IDIF-DA were similar, with an average relative error of 9%. The mean Pearson correlations between kinetic parameters (K₁, k₂, k₃) estimated with IDIF-DA and IDIF-Brain for each voxel were between 0.92 and 0.99 in all subjects, and for K_i, it was above 0.97. This study introduces a new approach for AIF estimation in dynamic PET using brain PET images, a probabilistic vascular atlas, and machine learning techniques. The findings demonstrate the feasibility of non-invasive and subject-specific AIF estimation for SAFOV scanners.

Medical image super-resolution (SR) is a classic regression task in low-level vision. Limited by hardware limitations, acquisition time, low radiation dose, and other factors, the spatial resolution of some medical images is not sufficient. To address this problem, many different SR methods have been proposed. Especially in recent years, medical image SR networks based on deep learning have been vigorously developed. This survey provides a modular and detailed introduction to the key components of medical image SR technology based on deep learning. In this paper, we first introduce the background concepts of deep learning and medical image SR task. Subsequently, we present a comprehensive analysis of the key components from the perspectives of effective architecture, upsampling module, learning strategy, and image quality assessment of medical image SR networks. Furthermore, we focus on the urgent problems that need to be addressed in the medical image SR task based on deep learning. And finally we summarize the trends and challenges of future development.

<b>Background and purpose:</b> This study focuses on an objective evaluation of a novel reconstruction algorithm-Deep Learning Image Reconstruction (DLIR)-ability to improve image quality and reduce radiation dose compared to the established standard of Adaptive Statistical Iterative Reconstruction-V (ASIR-V), in unenhanced head computed tomography (CT). <b>Materials and methods:</b> A retrospective analysis of 163 consecutive unenhanced head CTs was conducted. Image quality assessment was computed on the objective parameters of Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR), derived from 5 regions of interest (ROI). The evaluation of DLIR dose reduction abilities was based on the analysis of the PACS derived parameters of dose length product and computed tomography dose index volume (CTDIvol). <b>Results:</b> Following the application of rigorous criteria, the study comprised 35 patients. Significant image quality improvement was achieved with the implementation of DLIR, as evidenced by up to a 145% and 160% increase in SNR in supra- and infratentorial regions, respectively. CNR measurements further confirmed the superiority of DLIR over ASIR-V, with an increase of 171.5% in the supratentorial region and a 59.3% increase in the infratentorial region. Despite the signal improvement and noise reduction DLIR facilitated radiation dose reduction of up to 44% in CTDIvol. <b>Conclusion:</b> Implementation of DLIR in head CT scans enables significant image quality improvement and dose reduction abilities compared to standard ASIR-V. However, the dose reduction feature was proven insufficient to counteract the lack of gantry angulation in wide-detector scanners.

High-quality 3D fetal brain MRI reconstruction from motion-corrupted 2D slices is crucial for clinical diagnosis. Reliable slice-to-volume registration (SVR)-based motion correction and super-resolution reconstruction (SRR) methods are essential. Deep learning (DL) has demonstrated potential in enhancing SVR and SRR when compared to conventional methods. However, it requires large-scale external training datasets, which are difficult to obtain for clinical fetal MRI. To address this issue, we propose an unsupervised iterative SVR-SRR framework for isotropic HR volume reconstruction. Specifically, SVR is formulated as a function mapping a 2D slice and a 3D target volume to a rigid transformation matrix, which aligns the slice to the underlying location in the target volume. The function is parameterized by a convolutional neural network, which is trained by minimizing the difference between the volume slicing at the predicted position and the input slice. In SRR, a decoding network embedded within a deep image prior framework is incorporated with a comprehensive image degradation model to produce the high-resolution (HR) volume. The deep image prior framework offers a local consistency prior to guide the reconstruction of HR volumes. By performing a forward degradation model, the HR volume is optimized by minimizing loss between predicted slices and the observed slices. Comprehensive experiments conducted on large-magnitude motion-corrupted simulation data and clinical data demonstrate the superior performance of the proposed framework over state-of-the-art fetal brain reconstruction frameworks.

Medical image reconstruction from measurement data is a vital but challenging inverse problem. Deep learning approaches have achieved promising results, but often requires paired measurement and high-quality images, which is typically simulated through a forward model, i.e., retrospective reconstruction. However, training on simulated pairs commonly leads to performance degradation on real prospective data due to the retrospective-to-prospective gap caused by incomplete imaging knowledge in simulation. To address this challenge, this paper introduces imaging Knowledge-Informed Dynamic Optimal Transport (KIDOT), a novel dynamic optimal transport framework with optimality in the sense of preserving consistency with imaging physics in transport, that conceptualizes reconstruction as finding a dynamic transport path. KIDOT learns from unpaired data by modeling reconstruction as a continuous evolution path from measurements to images, guided by an imaging knowledge-informed cost function and transport equation. This dynamic and knowledge-aware approach enhances robustness and better leverages unpaired data while respecting acquisition physics. Theoretically, we demonstrate that KIDOT naturally generalizes dynamic optimal transport, ensuring its mathematical rationale and solution existence. Extensive experiments on MRI and CT reconstruction demonstrate KIDOT's superior performance.

Functional ultrasound (fUS) imaging provides exceptional spatiotemporal resolution for neurovascular mapping, yet its practical application is significantly hampered by critical challenges. Foremost among these are data scarcity, arising from ethical considerations and signal degradation through the cranium, which collectively limit dataset diversity and compromise the fairness of downstream machine learning models.

<i>Objective.</i>Magnetic resonance imaging (MRI) is critical in medical diagnosis and treatment by capturing detailed features, such as subtle tissue changes, which help clinicians make precise diagnoses. However, the widely used single diffusion model has limitations in accurately capturing more complex details. This study aims to address these limitations by proposing an efficient method to enhance the reconstruction of detailed features in MRI.<i>Approach.</i>We present a detail-preserving reconstruction method that leverages multiple diffusion models (DP-MDM) to extract structural and detailed features in the k-space domain, which complements the image domain. Since high-frequency information in k-space is more systematically distributed around the periphery compared to the irregular distribution of detailed features in the image domain, this systematic distribution allows for more efficient extraction of detailed features. To further reduce redundancy and enhance model performance, we introduce virtual binary masks with adjustable circular center windows that selectively focus on high-frequency regions. These masks align with the frequency distribution of k-space data, enabling the model to focus more efficiently on high-frequency information. The proposed method employs a cascaded architecture, where the first diffusion model recovers low-frequency structural components, with subsequent models enhancing high-frequency details during the iterative reconstruction stage.<i>Main results.</i>Experimental results demonstrate that DP-MDM achieves superior performance across multiple datasets. On the<i>T1-GE brain</i>dataset with 2D random sampling at<i>R</i>= 15, DP-MDM achieved 35.14 dB peak signal-to-noise ratio (PSNR) and 0.8891 structural similarity (SSIM), outperforming other methods. The proposed method also showed robust performance on the<i>Fast-MRI</i>and<i>Cardiac MR</i>datasets, achieving the highest PSNR and SSIM values.<i>Significance.</i>DP-MDM significantly advances MRI reconstruction by balancing structural integrity and detail preservation. It not only enhances diagnostic accuracy through improved image quality but also offers a versatile framework that can potentially be extended to other imaging modalities, thereby broadening its clinical applicability.

Volumetric rendering of Computed Tomography (CT) scans is crucial for visualizing complex 3D anatomical structures in medical imaging. Current high-fidelity approaches, especially neural rendering techniques, require time-consuming per-scene optimization, limiting clinical applicability due to computational demands and poor generalizability. We propose Render-FM, a novel foundation model for direct, real-time volumetric rendering of CT scans. Render-FM employs an encoder-decoder architecture that directly regresses 6D Gaussian Splatting (6DGS) parameters from CT volumes, eliminating per-scan optimization through large-scale pre-training on diverse medical data. By integrating robust feature extraction with the expressive power of 6DGS, our approach efficiently generates high-quality, real-time interactive 3D visualizations across diverse clinical CT data. Experiments demonstrate that Render-FM achieves visual fidelity comparable or superior to specialized per-scan methods while drastically reducing preparation time from nearly an hour to seconds for a single inference step. This advancement enables seamless integration into real-time surgical planning and diagnostic workflows. The project page is: https://gaozhongpai.github.io/renderfm/.

High-resolution magnetic resonance angiography (~ 50 μm³ MRA) data plays a critical role in the accurate diagnosis of various vascular disorders. However, it is very challenging to acquire, and it is susceptible to artifacts and noise which limits its ability to visualize smaller blood vessels and necessitates substantial noise reduction measures. Among many techniques, the BM4D filter is a state-of-the-art denoising technique but comes with high computational cost, particularly for high-resolution 3D MRA data. In this research, five different optimized convolutional neural networks were utilized to denoise contrast-enhanced UTE-MRA data using a supervised learning approach. Since noise-free MRA data is challenging to acquire, the denoised image using BM4D filter was used as ground truth and this research mainly focused on reducing computational cost and inference time for denoising high-resolution UTE-MRA data. All five models were able to generate nearly similar denoised data compared to the ground truth with different computational footprints. Among all, the nested-UNet model generated almost similar images with the ground truth and achieved SSIM, PSNR, and MSE of 0.998, 46.12, and 3.38e-5 with 3× faster inference time than the BM4D filter. In addition, most optimized models like UNet and attention-UNet models generated nearly similar images with nested-UNet but 8.8× and 7.1× faster than the BM4D filter. In conclusion, using highly optimized networks, we have shown the possibility of denoising high-resolution UTE-MRA data with significantly shorter inference time, even with limited datasets from animal models. This can potentially make high-resolution 3D UTE-MRA data to be less computationally burdensome.

CT imaging provides essential information about internal anatomy; however, conventional CT imaging delivers radiation doses that can become problematic for patients requiring repeated imaging, highlighting the need for dose-reduction techniques. This study aims to reduce radiation doses without compromising image quality. We propose an approach that combines Neural Attenuation Fields (NAF) with an active learning strategy to better optimize CT reconstructions given a limited number of X-ray projections. Our method uses a secondary neural network to predict the Peak Signal-to-Noise Ratio (PSNR) of 2D projections generated by NAF from a range of angles in the operational range of the CT scanner. This prediction serves as a guide for the active learning process in choosing the most informative projections. In contrast to conventional techniques that acquire all X-ray projections in a single session, our technique iteratively acquires projections. The iterative process improves reconstruction quality, reduces the number of required projections, and decreases patient radiation exposure. We tested our methodology on spinal imaging using a limited subset of the VerSe 2020 dataset. We compare image quality metrics (PSNR3D, SSIM3D, and PSNR2D) to the baseline method and find significant improvements. Our method achieves the same quality with 36 projections as the baseline method achieves with 60. Our findings demonstrate that our approach achieves high-quality 3D CT reconstructions from sparse data, producing clearer and more detailed images of anatomical structures. This work lays the groundwork for advanced imaging techniques, paving the way for safer and more efficient medical imaging procedures.