Developing and Validating Multi-Modal Models for Mortality Prediction in COVID-19 Patients: a Multi-center Retrospective Study.

The unprecedented global crisis brought about by the COVID-19 pandemic has sparked numerous efforts to create predictive models for the detection and prognostication of SARS-CoV-2 infections with the goal of helping health systems allocate resources. Machine learning models, in particular, hold promise for their ability to leverage patient clinical information and medical images for prediction. However, most of the published COVID-19 prediction models thus far have little clinical utility due to methodological flaws and lack of appropriate validation. In this paper, we describe our methodology to develop and validate multi-modal models for COVID-19 mortality prediction using multi-center patient data. The models for COVID-19 mortality prediction were developed using retrospective data from Madrid, Spain (N = 2547) and were externally validated in patient cohorts from a community hospital in New Jersey, USA (N = 242) and an academic center in Seoul, Republic of Korea (N = 336). The models we developed performed differently across various clinical settings, underscoring the need for a guided strategy when employing machine learning for clinical decision-making. We demonstrated that using features from both the structured electronic health records and chest X-ray imaging data resulted in better 30-day mortality prediction performance across all three datasets (areas under the receiver operating characteristic curves: 0.85 (95% confidence interval: 0.83-0.87), 0.76 (0.70-0.82), and 0.95 (0.92-0.98)). We discuss the rationale for the decisions made at every step in developing the models and have made our code available to the research community. We employed the best machine learning practices for clinical model development. Our goal is to create a toolkit that would assist investigators and organizations in building multi-modal models for prediction, classification, and/or optimization.

On the R2⁎ relaxometry in complex multi-peak multi-Echo chemical shift-based water-fat quantification: Applications to the neuromuscular diseases.

PURPOSE: Investigation of the feasibility of the R2⁎ mapping techniques by using latest theoretical models corrected for confounding factors and optimized for signal to noise ratio. THEORY AND METHODS: The improvement of the performance of state of the art magnetic resonance imaging (MRI) relaxometry algorithms is challenging because of a non-negligible bias and still unresolved numerical instabilities. Here, R2⁎ mapping reconstructions, including complex fitting with multi-spectral fat-correction by using single-decay and double-decay formulation, are deeply studied in order to investigate and identify optimal configuration parameters and minimize the occurrence of numerical artifacts. The effects of echo number, echo spacing, and fat/water relaxation model type are evaluated through both simulated and in-vivo data. We also explore the stability and feasibility of the fat/water relaxation model by analyzing the impact of high percentage of fat infiltrations and local transverse relaxation differences among biological species. RESULTS: The main limits of the MRI relaxometry are the presence of bias and the occurrence of artifacts, which significantly affect its accuracy. Chemical-shift complex R2⁎-correct single-decay reconstructions exhibit a large bias in presence of a significant difference in the relaxation rates of fat and water and with fat concentration larger than 30%. We find that for fat-dominated tissues or in patients affected by extensive iron deposition, MRI reconstructions accounting for multi-exponential relaxation time provide accurate R2⁎ measurements and are less prone to numerical artifacts. CONCLUSIONS: Complex fitting and fat-correction with multi-exponential decay formulation outperforms the conventional single-decay approximation in various diagnostic scenarios. Although it still lacks of numerical stability, which requires model enhancement and support from spectroscopy, it offers promising perspectives for the development of relaxometry as a reliable tool to improve tissue characterization and monitoring of neuromuscular disorders.

Basal ganglia network by constrained spherical deconvolution: a possible cortico-pallidal pathway?

In the recent past, basal ganglia circuitry was simplified as represented by the direct and indirect pathways and by hyperdirect pathways. Based on data from animal studies, we hypothesized a fourth pathway, the cortico-pallidal, pathway, that complements the hyperdirect pathway to the subthalamus. Ten normal brains were analyzed by using the high angular resolution diffusion imaging-constrained spherical deconvolution (CSD)-based technique. The study was performed with a 3T magnetic resonance imaging (MRI) scanner (Achieva, Philips Healthcare, Best, Netherlands); by using a 32-channel SENSE head coil. We showed that CSD is a powerful technique that allows a fine evaluation of both the long and small tracts between cortex and basal ganglia, including direct, indirect, and hyperdirect pathways. In addition, a pathway directly connecting the cortex to the globus pallidus was seen. Our results confirm that the CSD tractography is a valuable technique allowing a reliable reconstruction of small- and long-fiber pathways in brain regions with multiple fiber orientations, such as basal ganglia. This could open a future scenario in which CSD could be used to focally target with deep brain stimulation (DBS) the small bundles within the basal ganglia loops.

Cortical and subcortical connections of the human claustrum revealed in vivo by constrained spherical deconvolution tractography.

The claustrum is a thin layer of gray matter that is at the center of an active scientific debate. Recently, Constrained Spherical Deconvolution (CSD) tractography has proved to be an extraordinary tool allowing to track white matter fibers from cortex to cortical and subcortical targets with subvoxel resolution. The aim of this study was to evaluate claustral connectivity in the human brain. Ten normal brains were analyzed by using the High Angular Resolution Diffusion Imaging CSD-based technique. Tractography revealed 4 groups of white matter fibers connecting the claustrum with the brain cortex: Anterior, posterior, superior, and lateral. The anterior and posterior cortico-claustral tracts connected the claustrum to prefrontal cortex and visual areas. The superior tract linked the claustrum with sensory-motor areas, while the lateral pathway connected the claustrum to the auditory cortex. In addition, we demonstrated a claustral medial pathway connecting the claustrum with the basal ganglia, specifically with caudate nucleus, putamen, and globus pallidus. An interesting and exciting new finding was the demonstration of a bilateral connection between claustrum and contralateral cortical areas and a well-represented interclaustral communication with interconnection bundles interspersed within the bulk of the trunk of the corpus callosum. The physiological and pathophysiological relevance of these findings are discussed.

MRI 3D lateral cerebral ventricles in living humans: morphological and morphometrical age-, gender-related preliminary study.

Morphological and volumetric variabilities of lateral ventricles are considered indirect indicators of age-and gender-related reductions of white and gray matter. However, no studies have classified lateral ventricles with different morphologies or showed its asymmetric shapes in healthy subjects. We performed an analysis on living subjects, using 3D volume rendering techniques. Eighty-five healthy Caucasian volunteers (49 women and 36 men aged 19-69 years) were scanned by a Philips Achieva 3T R2.6. Three-dimensional reconstruction allowed us to identify three main morphological shapes in living subjects and to show asymmetries between horns. We also assessed the surface deformation of the cerebral ventricles to identify region-specific shape differences in aging healthy adults. Statistical analysis showed significant gender- and age-related volume differences. An increase in lateral ventricle volume appears to be a constant, linear function of age throughout adult life.