RESUMEN
Although fully automated volumetric approaches for monitoring brain tumor response have many advantages, most available deep learning models are optimized for highly curated, multi-contrast MRI from newly diagnosed gliomas, which are not representative of post-treatment cases in the clinic. Improving segmentation for treated patients is critical to accurately tracking changes in response to therapy. We investigated mixing data from newly diagnosed (n = 208) and treated (n = 221) gliomas in training, applying transfer learning (TL) from pre- to post-treatment imaging domains, and incorporating spatial regularization for T2-lesion segmentation using only T2 FLAIR images as input to improve generalization post-treatment. These approaches were evaluated on 24 patients suspected of progression who had received prior treatment. Including 26% of treated patients in training improved performance by 13.9%, and including more treated and untreated patients resulted in minimal changes. Fine-tuning with treated glioma improved sensitivity compared to data mixing by 2.5% (p < 0.05), and spatial regularization further improved performance when used with TL by 95th HD, Dice, and sensitivity (6.8%, 0.8%, 2.2%; p < 0.05). While training with ≥60 treated patients yielded the majority of performance gain, TL and spatial regularization further improved T2-lesion segmentation to treated gliomas using a single MR contrast and minimal processing, demonstrating clinical utility in response assessment.
RESUMEN
A system for analysis of histopathology data within a pharmaceutical R&D environment has been developed with the intention of enabling interdisciplinary collaboration. State-of-the-art AI tools have been deployed as easy-to-use self-service modules within an open-source whole slide image viewing platform, so that non-data scientist users (e.g., clinicians) can utilize and evaluate pre-trained algorithms and retrieve quantitative results. The outputs of analysis are automatically cataloged in the database to track data provenance and can be viewed interactively on the slide as annotations or heatmaps. Commonly used models for analysis of whole slide images including segmentation, extraction of hand-engineered features for segmented regions, and slide-level classification using multi-instance learning are included and new models can be added as needed. The source code that supports running inference with these models internally is backed up by a robust CI/CD pipeline to ensure model versioning, robust testing, and seamless deployment of the latest models. Examples of the use of this system in a pharmaceutical development workflow include glomeruli segmentation, enumeration of podocyte count from WT-1 immuno-histochemistry, measurement of beta-1 integrin target engagement from immunofluorescence, digital glomerular phenotyping from periodic acid-Schiff histology, PD-L1 score prediction using multi-instance learning, and the deployment of the open-source Segment Anything model to speed up annotation.
RESUMEN
Background: Monitoring lower-grade gliomas (LrGGs) for disease progression is made difficult by the limits of anatomical MRI to distinguish treatment related tissue changes from tumor progression. MR spectroscopic imaging (MRSI) offers additional metabolic information that can help address these challenges. The goal of this study was to compare longitudinal changes in multiparametric MRI, including diffusion weighted imaging, perfusion imaging, and 3D MRSI, for LrGG patients who progressed at the final time-point and those who remained clinically stable. Methods: Forty-one patients with LrGG who were clinically stable were longitudinally assessed for progression. Changes in anatomical, diffusion, perfusion and MRSI data were acquired and compared between patients who remained clinically stable and those who progressed. Results: Thirty-one patients remained stable, and 10 patients progressed. Over the study period, progressed patients had a significantly greater increase in normalized choline, choline-to-N-acetylaspartic acid index (CNI), normalized creatine, and creatine-to-N-acetylaspartic acid index (CRNI), than stable patients. CRNI was significantly associated with progression status and WHO type. Progressed astrocytoma patients had greater increases in CRNI than stable astrocytoma patients. Conclusions: LrGG patients in surveillance with tumors that progressed had significantly increasing choline and creatine metabolite signals on MRSI, with a trend of increasing T2 FLAIR volumes, compared to LrGG patients who remained stable. These data show that MRSI can be used in conjunction with anatomical imaging studies to gain a clearer picture of LrGG progression, especially in the setting of clinical ambiguity.
RESUMEN
Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.
Asunto(s)
COVID-19/fisiopatología , Aprendizaje Automático , Evaluación de Resultado en la Atención de Salud , COVID-19/terapia , COVID-19/virología , Registros Electrónicos de Salud , Humanos , Pronóstico , SARS-CoV-2/aislamiento & purificaciónRESUMEN
How do functional brain networks emerge from the underlying wiring of the brain? We examine how resting-state functional activation patterns emerge from the underlying connectivity and length of white matter fibers that constitute its "structural connectome". By introducing realistic signal transmission delays along fiber projections, we obtain a complex-valued graph Laplacian matrix that depends on two parameters: coupling strength and oscillation frequency. This complex Laplacian admits a complex-valued eigen-basis in the frequency domain that is highly tunable and capable of reproducing the spatial patterns of canonical functional networks without requiring any detailed neural activity modeling. Specific canonical functional networks can be predicted using linear superposition of small subsets of complex eigenmodes. Using a novel parameter inference procedure we show that the complex Laplacian outperforms the real-valued Laplacian in predicting functional networks. The complex Laplacian eigenmodes therefore constitute a tunable yet parsimonious substrate on which a rich repertoire of realistic functional patterns can emerge. Although brain activity is governed by highly complex nonlinear processes and dense connections, our work suggests that simple extensions of linear models to the complex domain effectively approximate rich macroscopic spatial patterns observable on BOLD fMRI.
Asunto(s)
Encéfalo/anatomía & histología , Encéfalo/fisiología , Imagen por Resonancia Magnética , Modelos Teóricos , Red Nerviosa/anatomía & histología , Red Nerviosa/fisiología , Neuroimagen , Encéfalo/diagnóstico por imagen , Red en Modo Predeterminado/anatomía & histología , Red en Modo Predeterminado/diagnóstico por imagen , Red en Modo Predeterminado/fisiología , Humanos , Red Nerviosa/diagnóstico por imagenRESUMEN
The rigid constraints of chemistry-dictated by quantum mechanics and the discrete nature of the atom-limit the set of observable atomic crystal structures. What structures are possible in the absence of these constraints? Here, we systematically crystallize one-component systems of particles interacting with isotropic multiwell pair potentials. We investigate two tunable families of pairwise interaction potentials. Our simulations self-assemble a multitude of crystal structures ranging from basic lattices to complex networks. Sixteen of the structures have natural analogs spanning all coordination numbers found in inorganic chemistry. Fifteen more are hitherto unknown and occupy the space between covalent and metallic coordination environments. The discovered crystal structures constitute targets for self-assembly and expand our understanding of what a crystal structure can look like.
RESUMEN
'Federated Learning' (FL) is a method to train Artificial Intelligence (AI) models with data from multiple sources while maintaining anonymity of the data thus removing many barriers to data sharing. During the SARS-COV-2 pandemic, 20 institutes collaborated on a healthcare FL study to predict future oxygen requirements of infected patients using inputs of vital signs, laboratory data, and chest x-rays, constituting the "EXAM" (EMR CXR AI Model) model. EXAM achieved an average Area Under the Curve (AUC) of over 0.92, an average improvement of 16%, and a 38% increase in generalisability over local models. The FL paradigm was successfully applied to facilitate a rapid data science collaboration without data exchange, resulting in a model that generalised across heterogeneous, unharmonized datasets. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as set the stage for broader use of FL in healthcare.
RESUMEN
Background: Bone marrow fat (BMF) fraction quantification in vertebral bodies is used as a novel imaging biomarker to assess and characterize chronic lower back pain. However, manual segmentation of vertebral bodies is time consuming and laborious. Purpose: (1) Develop a deep learning pipeline for segmentation of vertebral bodies using quantitative water-fat MRI. (2) Compare BMF measurements between manual and automatic segmentation methods to assess performance. Materials and Methods: In this retrospective study, MR images using a 3D spoiled gradient-recalled echo (SPGR) sequence with Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation (IDEAL) reconstruction algorithm were obtained in 57 subjects (28 women, 29 men, mean age, 47.2 ± 12.6 years). An artificial network was trained for 100 epochs on a total of 165 lumbar vertebrae manually segmented from 31 subjects. Performance was assessed by analyzing the receiver operating characteristic curve, precision-recall, F1 scores, specificity, sensitivity, and similarity metrics. Bland-Altman analysis was used to assess performance of BMF fraction quantification using the predicted segmentations. Results: The deep learning segmentation method achieved an AUC of 0.92 (CI 95%: 0.9186, 0.9195) on a testing dataset (n = 24 subjects) on classification of pixels as vertebrae. A sensitivity of 0.99 and specificity of 0.80 were achieved for a testing dataset, and a mean Dice similarity coefficient of 0.849 ± 0.091. Comparing manual and automatic segmentations on fat fraction maps of lumbar vertebrae (n = 124 vertebral bodies) using Bland-Altman analysis resulted in a bias of only -0.605% (CI 95% = -0.847 to -0.363%) and agreement limits of -3.275% and +2.065%. Automatic segmentation was also feasible in 16 ± 1 s. Conclusion: Our results have demonstrated the feasibility of automated segmentation of vertebral bodies using deep learning models on water-fat MR (Dixon) images to define vertebral regions of interest with high specificity. These regions of interest can then be used to quantify BMF with comparable results as manual segmentation, providing a framework for completely automated investigation of vertebral changes in CLBP.
Asunto(s)
Tejido Adiposo/diagnóstico por imagen , Médula Ósea/diagnóstico por imagen , Aprendizaje Profundo , Columna Vertebral/diagnóstico por imagen , Adulto , Femenino , Humanos , Procesamiento de Imagen Asistido por Computador , Imagen por Resonancia Magnética/métodos , Masculino , Persona de Mediana Edad , Estudios RetrospectivosRESUMEN
BACKGROUND: The release of a broad, longitudinal anatomical dataset by the Parkinson's Progression Markers Initiative promoted a surge of machine-learning studies aimed at predicting disease onset and progression. However, the excessive number of features used in these models often conceals their relationship to the Parkinsonian symptomatology. OBJECTIVES: The aim of this study is two-fold: (i) to predict future motor and cognitive impairments up to four years from brain features acquired at baseline; and (ii) to interpret the role of pivotal brain regions responsible for different symptoms from a neurological viewpoint. METHODS: We test several deep-learning neural network configurations, and report our best results obtained with an autoencoder deep-learning model, run on a 5-fold cross-validation set. Comparison with Existing Methods: Our approach improves upon results from standard regression and others. It also includes neuroimaging biomarkers as features. RESULTS: The relative contributions of pivotal brain regions to each impairment change over time, suggesting a dynamical reordering of culprits as the disease progresses. Specifically, the Putamen is initially the most critical region accounting for the overall cognitive state, only being surpassed by the Substantia Nigra in later years. The Pallidum is the first region to influence motor scores, followed by the parahippocampal and ambient gyri, and the anterior orbital gyrus. CONCLUSIONS: While the causal link between regional brain atrophy and Parkinson symptomatology is poorly understood, our methods demonstrate that the contributions of pivotal regions to cognitive and motor impairments are more dynamical than generally appreciated.
RESUMEN
Low-dimensional objects such as molecular strands, ladders, and sheets have intrinsic features that affect their propensity to fold into 3D objects. Understanding this relationship remains a challenge for de novo design of functional structures. Using molecular dynamics simulations, we investigate the refolding of the 24 possible 2D unfoldings ("nets") of the three simplest Platonic shapes and demonstrate that attributes of a net's topology-net compactness and leaves on the cutting graph-correlate with thermodynamic folding propensity. To explain these correlations we exhaustively enumerate the pathways followed by nets during folding and identify a crossover temperature [Formula: see text] below which nets fold via nonnative contacts (bonds must break before the net can fold completely) and above which nets fold via native contacts (newly formed bonds are also present in the folded structure). Folding above [Formula: see text] shows a universal balance between reduction of entropy via the elimination of internal degrees of freedom when bonds are formed and gain in potential energy via local, cooperative edge binding. Exploiting this universality, we devised a numerical method to efficiently compute all high-temperature folding pathways for any net, allowing us to predict, among the combined 86,760 nets for the remaining Platonic solids, those with highest folding propensity. Our results provide a general heuristic for the design of 2D objects to stochastically fold into target 3D geometries and suggest a mechanism by which geometry and folding propensity are related above [Formula: see text], where native bonds dominate folding.
RESUMEN
Viscoelastic properties are central for gels and other materials. Simultaneously, high storage and loss moduli are difficult to attain due to their contrarian requirements to chemical structure. Biomimetic inorganic nanoparticles offer a promising toolbox for multiscale engineering of gel mechanics, but a conceptual framework for their molecular, nanoscale, mesoscale, and microscale engineering as viscoelastic materials is absent. Here we show nanoparticle gels with simultaneously high storage and loss moduli from CdTe nanoparticles. Viscoelastic figure of merit reaches 1.83 MPa exceeding that of comparable gels by 100-1000 times for glutathione-stabilized nanoparticles. The gels made from the smallest nanoparticles display the highest stiffness, which was attributed to the drastic change of GSH configurations when nanoparticles decrease in size. A computational model accounting for the difference in nanoparticle interactions for variable GSH configurations describes the unusual trends of nanoparticle gel viscoelasticity. These observations are generalizable to other NP gels interconnected by supramolecular interactions and lead to materials with high-load bearing abilities and energy dissipation needed for multiple technologies.
Asunto(s)
Hidrogeles/síntesis química , Nanopartículas/química , Materiales Biomiméticos , Compuestos de Cadmio/química , Glutatión/química , Fenómenos Mecánicos , Telurio/química , Sustancias ViscoelásticasRESUMEN
Quasicrystals are frequently encountered in condensed matter. They are important candidates for equilibrium phases from the atomic scale to the nanoscale. Here, we investigate the computational self-assembly of four quasicrystals in a single model system of identical particles interacting with a tunable isotropic pair potential. We reproduce a known icosahedral quasicrystal and report a decagonal quasicrystal, a dodecagonal quasicrystal, and an octagonal quasicrystal. The quasicrystals have low coordination number or occur in systems with mesoscale density variations. We also report a network gel phase.
RESUMEN
We study photonic band gap formation in two-dimensional high-refractive-index disordered materials where the dielectric structure is derived from packing disks in real and reciprocal space. Numerical calculations of the photonic density of states demonstrate the presence of a band gap for all polarizations in both cases. We find that the band gap width is controlled by the increase in positional correlation inducing short-range order and hyperuniformity concurrently. Our findings suggest that the optimization of short-range order, in particular the tailoring of Bragg scattering at the isotropic Brillouin zone, are of key importance for designing disordered PBG materials.
RESUMEN
Colloidal crystal structures with complexity and diversity rivaling atomic and molecular crystals have been predicted and obtained for hard particles by entropy maximization. However, thus far homochiral colloidal crystals, which are candidates for photonic metamaterials, are absent. Using Monte Carlo simulations we show that chiral polyhedra exhibiting weak directional entropic forces self-assemble either an achiral crystal or a chiral crystal with limited control over the crystal handedness. Building blocks with stronger faceting exhibit higher selectivity and assemble a chiral crystal with handedness uniquely determined by the particle chirality. Tuning the strength of directional entropic forces by means of particle rounding or the use of depletants allows for reconfiguration between achiral and homochiral crystals. We rationalize our findings by quantifying the chirality strength of each particle, both from particle geometry and potential of mean force and torque diagrams.
Asunto(s)
Coloides/química , Modelos Químicos , Cristalización , Entropía , Método de Montecarlo , EstereoisomerismoRESUMEN
Efforts to impart elasticity and multifunctionality in nanocomposites focus mainly on integrating polymeric and nanoscale components. Yet owing to the stochastic emergence and distribution of strain-concentrating defects and to the stiffening of nanoscale components at high strains, such composites often possess unpredictable strain-property relationships. Here, by taking inspiration from kirigamithe Japanese art of paper cuttingwe show that a network of notches made in rigid nanocomposite and other composite sheets by top-down patterning techniques prevents unpredictable local failure and increases the ultimate strain of the sheets from 4 to 370%. We also show that the sheets' tensile behaviour can be accurately predicted through finite-element modelling. Moreover, in marked contrast to other stretchable conductors, the electrical conductance of the stretchable kirigami sheets is maintained over the entire strain regime, and we demonstrate their use to tune plasma-discharge phenomena. The unique properties of kirigami nanocomposites as plasma electrodes open up a wide range of novel technological solutions for stretchable electronics and optoelectronic devices, among other application possibilities.
Asunto(s)
Nanocompuestos/química , Nanocompuestos/ultraestructura , Ingeniería Química/métodos , Elasticidad , Conductividad Eléctrica , Análisis de Elementos Finitos , Microscopía Electrónica de Rastreo , Nanotecnología/métodos , Nanotubos de Carbono/química , Nanotubos de Carbono/ultraestructura , Impresión Tridimensional , Estrés MecánicoRESUMEN
The relationship between colloidal building blocks and their assemblies is an active field of research. As a strategy for targeting novel crystal structures, we examine the use of Voronoi particles, which are hard, space-filling particles in the shape of Voronoi cells of a target structure. Although Voronoi particles stabilize their target structure in the limit of high pressure by construction, the thermodynamic assembly of the same structure at moderate pressure, close to the onset of crystallization, is not guaranteed. Indeed, we find that a more symmetric crystal is often preferred due to additional entropic contributions arising from configurational or occupational degeneracy. We characterize the assembly behavior of the Voronoi particles in terms of the symmetries of the building blocks as well as the symmetries of crystal structures and demonstrate how controlling the degeneracies through a modification of particle shape and field-directed assembly can significantly improve the assembly propensity.
RESUMEN
Icosahedral quasicrystals (IQCs) are a form of matter that is ordered but not periodic in any direction. All reported IQCs are intermetallic compounds and either of face-centred-icosahedral or primitive-icosahedral type, and the positions of their atoms have been resolved from diffraction data. However, unlike axially symmetric quasicrystals, IQCs have not been observed in non-atomic (that is, micellar or nanoparticle) systems, where real-space information would be directly available. Here, we show that an IQC can be assembled by means of molecular dynamics simulations from a one-component system of particles interacting via a tunable, isotropic pair potential extending only to the third-neighbour shell. The IQC is body-centred, self-assembles from a fluid phase, and in parameter space neighbours clathrates and other tetrahedrally bonded crystals. Our findings elucidate the structure and dynamics of the IQC, and suggest routes to search for it and design it in soft matter and nanoscale systems.
RESUMEN
Predicting structure from the attributes of a material's building blocks remains a challenge and central goal for materials science. Isolating the role of building block shape for self-assembly provides insight into the ordering of molecules and the crystallization of colloids, nanoparticles, proteins, and viruses. We investigated 145 convex polyhedra whose assembly arises solely from their anisotropic shape. Our results demonstrate a remarkably high propensity for thermodynamic self-assembly and structural diversity. We show that from simple measures of particle shape and local order in the fluid, the assembly of a given shape into a liquid crystal, plastic crystal, or crystal can be predicted.
RESUMEN
Polyhedra and their arrangements have intrigued humankind since the ancient Greeks and are today important motifs in condensed matter, with application to many classes of liquids and solids. Yet, little is known about the thermodynamically stable phases of polyhedrally shaped building blocks, such as faceted nanoparticles and colloids. Although hard particles are known to organize due to entropy alone, and some unusual phases are reported in the literature, the role of entropic forces in connection with polyhedral shape is not well understood. Here, we study thermodynamic self-assembly of a family of truncated tetrahedra and report several atomic crystal isostructures, including diamond, ß-tin, and high-pressure lithium, as the polyhedron shape varies from tetrahedral to octahedral. We compare our findings with the densest packings of the truncated tetrahedron family obtained by numerical compression and report a new space-filling polyhedron, which has been overlooked in previous searches. Interestingly, the self-assembled structures differ from the densest packings. We show that the self-assembled crystal structures can be understood as a tendency for polyhedra to maximize face-to-face alignment, which can be generalized as directional entropic forces.