Carlos Alberola-López

High-performance medical segmentation models are often benchmarked on high-end GPUs. Such benchmarks do not provide useful performance insights for point-of-care low-end devices. This work, firstly, posits that to achieve improved clinical adoption of AI-powered segmentation models, especially in reduced manpower settings like rural hospitals, we need benchmarks that provide actionable insights on the degree to which high-performance models address five deployment constraints viz: resource-effectiveness for low-end computing devices, clinically acceptable accuracy, clinically compatible execution times, localization of user data, and user-based finetuning. In this work, five state-of-the-art foundation segmentation models and one target-specific model were systematically evaluated on three multi-organ medical datasets. Furthermore, the best-ranking foundation model and target-specific model were benchmarked on three low-end devices. Our findings show that lightweight foundation models provided the best performance trade-off and are easily user-fine-tuned on custom datasets. Target-specific models provide high accuracy out-of-the-box, but may require significant optimisation to deliver comparably fast execution times and user-based finetuning on low-end devices. The methods and results from this research provide actionable insights on high-performance medical segmentation models for low-end computing devices, as a necessary step towards improved adoption in resource-limited clinical settings.

Purpose: First‐pass perfusion cardiovascular MR (FPP‐CMR) enables the non‐invasive diagnosis of microcirculation and coronary artery disease. In free‐breathing FPP‐CMR, motion correction is usually performed in the image domain, requiring an initial reconstruction. This fact hinders its use in model‐based and deep learning reconstructions, which present remarkable performance in obtaining high‐quality images from highly accelerated acquisitions. We address this challenge by estimating and correcting respiratory motion in free‐breathing FPP‐CMR directly in k ‐space. Methods: We propose K‐CC‐MoCo, an inter‐frame rigid motion correction approach formulated exclusively in k ‐space that handles dynamic contrast through a specifically targeted design of the normalized cross‐correlation (CC) objective function to deal with the dynamic contrast. In addition, an ROI‐based coil‐compression approach was employed to focus the optimization on the heart region. The proposed method was compared to state‐of‐the‐art image‐based registration using a digital phantom and real free‐breathing acquisitions with different accelerations. Results: The proposed k ‐space‐based method is approximately 2× faster and can correct respiratory motion even at high acceleration factors (up to 50×), where the image‐based method fails due to severe undersampling artifacts. Notably, after K‐CC‐MoCo, the time‐averaged images are visibly less blurred. Quantitative metrics (SSIM, etc.) support this conclusion. Conclusion: K‐CC‐MoCo outperforms image‐based correction in free‐breathing FPP‐CMR acquisitions accelerated up to 50×. Respiratory motion is estimated and corrected in k ‐space, enabling its use for model‐based and/or deep learning reconstructions from highly accelerated scans.

We present MRSeqStudio, a new all-in-one web-based tool for MRI sequence developmentand simulation, with the physics-based simulator KomaMRI running at the back-end and our own sequence designer at the front-end. It combines accessibility, interactivity and technical flexibility, within an environment suitable for both education and research. Our tool provides MR sequence design and simulation as a service , with no local installation needed by the user; alternatively, the code is publicly available on GitHub, for users who wish to deploy the application on their own server.

Purpose: To extend the KomaMRI simulator with motion capabilities that enable the simulation of both simple, parametrizable motion patterns and arbitrarily complex spin trajectories. Additionally, we introduce a novel file format for storing and sharing dynamic digital phantoms. Methods: The existing Phantom structure in KomaMRI has been extended with a new motion field, which preserves support for static phantoms and introduces the capability to define dynamic models, composed of either a single or an arbitrary set of elementary motions. Each motion entry is composed of an action , a time curve , and a spin span , allowing for modular definition and full parameter independence across motion components. An HDF5‐based phantom file format has been defined, along with two I/O functions within the simulator. Motion‐related MRI experiments have been conducted for both illustrative and comparative evaluation. Results: Illustrative experiments reveal motion‐related effects commonly observed in real MRI, such as time of flight and phase contrast. Comparative evaluations show strong qualitative and quantitative agreement with results reported by other simulation tools and with at least a three‐fold reduction in computation time. Conclusions: The proposed extension adds flexible motion modeling and simulation capabilities to KomaMRI. The proposed new functions allow for accurate and high temporal and spatial resolution, only limited by computational cost.

Purpose: (a) To design a methodology for drawing random samples of any Ensemble Average Propagator (EAP) (b) to modify the KomaMRI simulator to accommodate them as realistic spin movements to simulate diffusion MRI (dMRI) and (c) to compare these simulations with those based on the Diffusion Tensor (DT) model. Theory and Methods: The rejection method is used for random sampling of EAPs: starting from a probability law that is easily sampled, and whose density function wraps the target EAP, samples are accepted when they lie inside the targeted region. This is used to sample the EAP as described by Mean Apparent Propagator MRI (MAP‐MRI) and in Spherical Convolution (SC) based on Spherical Harmonics (SH). With this methodology, MAP‐MRI and SC representations are calculated over in‐vitro pig hearts images, and a simulation of a pulsed‐gradient spin echo (PGSE) dMRI sequence inside the myocardial wall is undertaken with the KomaMRI simulator. Results: MAP‐MRI shows better agreement with the actual acquisition than conventional DT‐based simulations, in terms of Mean Squared Errors and correlation with improvements up to 1.7 % for the former and 2.2 % for the latter. Conclusion: dMRI sequences can be simulated accurately (yet, efficiently) if phantoms with a proper per‐spin description of the diffusion process are made available. Moreover, our findings suggest that the study of non‐Gaussian diffusion of the heart might be feasible, at least in vitro .

Background: Gadolinium-based contrast agents (GBCAs) are usually employed for glioma diagnosis. However, GBCAs raise safety concerns, lead to patient discomfort and increase costs. Parametric maps offer a potential solution by enabling quantification of subtle tissue changes without GBCAs, but they are not commonly used in clinical practice due to the need for specifically targeted sequences. This work proposes to predict post-contrast T1-weighted enhancement without GBCAs from pre-contrast conventional weighted images through synthetic parametric maps computed with generative artificial intelligence (deep learning). Methods: In this retrospective study, three datasets have been employed: (I) a proprietary dataset with 15 glioma patients (hereafter, GLIOMA dataset); (II) relaxometry maps from 5 healthy volunteers; and (III) UPenn-GBM, a public dataset with 493 glioblastoma patients. A deep learning method for synthesizing parametric maps from only two conventional weighted images is proposed. Particularly, we synthesize longitudinal relaxation time (T1), transversal relaxation time (T2), and proton density (PD) maps. The deep learning method is trained in a supervised manner with the GLIOMA dataset, which comprises weighted images and parametric maps obtained with magnetic resonance image compilation (MAGiC). Thus, MAGiC maps were used as references for the training. For testing, a leave-one-out scheme is followed. Finally, the synthesized maps are employed to predict T1-weighted enhancement without GBCAs. Our results are compared with those obtained by MAGiC; specifically, both the maps obtained with MAGiC and the synthesized maps are used to distinguish between healthy and abnormal tissue (ABN) and, particularly, tissues with and without T1-weighted enhancement. The generalization capability of the method was also tested on two additional datasets (healthy volunteers and the UPenn-GBM). Results: Parametric maps synthesized with deep learning obtained similar performance compared to MAGiC for discriminating normal from ABN (sensitivities: 88.37% vs. 89.35%) and tissue with and without T1-weighted enhancement (sensitivities: 93.26% vs. 87.29%) on the GLIOMA dataset. These values were comparable to those obtained on UPenn-GBM (sensitivities of 91.23% and 81.04% for each classification). Conclusions: Our results suggest the feasibility to predict T1-weighted-enhanced tissues from pre-contrast conventional weighted images using deep learning for the synthesis of parametric maps.

Cardiac CINE, a form of dynamic cardiac MRI, is indispensable in the diagnosis and treatment of heart conditions, offering detailed visualization essential for the early detection of cardiac diseases. As the demand for higher-resolution images increases, so does the volume of data requiring processing, presenting significant computational challenges that can impede the efficiency of diagnostic imaging. Our research presents an approach that takes advantage of the computational power of multiple Graphics Processing Units (GPUs) to address these challenges. GPUs are devices capable of performing large volumes of computations in a short period, and have significantly improved the cardiac MRI reconstruction process, allowing images to be produced faster. The innovation of our work resides in utilizing a multi-device system capable of processing the substantial data volumes demanded by high-resolution, five-dimensional cardiac MRI. This system surpasses the memory capacity limitations of single GPUs by partitioning large datasets into smaller, manageable segments for parallel processing, thereby preserving image integrity and accelerating reconstruction times. Utilizing OpenCL technology, our system offers adaptability and cross-platform functionality, ensuring wider applicability. The proposed multi-device approach offers an advancement in medical imaging, accelerating the reconstruction process and facilitating faster and more effective cardiac health assessment.