Lesion-preserving unpaired image-to-image translation between MRI and CT from ischemic stroke patients.

PURPOSE: Multiple medical imaging modalities are used for clinical follow-up ischemic stroke analysis. Mixed-modality datasets are challenging, both for clinical rating purposes and for training machine learning models. While image-to-image translation methods have been applied to harmonize stroke patient images to a single modality, they have only been used for paired data so far. In the more common unpaired scenario, the standard cycle-consistent generative adversarial network (CycleGAN) method is not able to translate the stroke lesions properly. Thus, the aim of this work was to develop and evaluate a novel image-to-image translation regularization approach for unpaired 3D follow-up stroke patient datasets. METHODS: A modified CycleGAN was used to translate images between 238 non-contrast computed tomography (NCCT) and 244 fluid-attenuated inversion recovery (FLAIR) MRI datasets, two of the most relevant follow-up modalities in clinical practice. We introduced an additional attention-guided mechanism to encourage an improved translation of the lesion and a gradient-consistency loss to preserve structural brain morphology. RESULTS: The proposed modifications were able to preserve the overall quality provided by the CycleGAN translation. This was confirmed by the FID score and gradient correlation results. Furthermore, the lesion preservation was significantly improved compared to a standard CycleGAN. This was evaluated for location and volume with segmentation models, which were trained on real datasets and applied to the translated test images. Here, the Dice score coefficient resulted in 0.81 and 0.62 for datasets translated to FLAIR and NCCT, respectively, compared to 0.57 and 0.50 for the corresponding datasets translated using a standard CycleGAN. Finally, an analysis of the distribution of mean lesion intensities showed substantial improvements. CONCLUSION: The results of this work show that the proposed image-to-image translation method is effective at preserving stroke lesions in unpaired modality translation, supporting its potential as a tool for stroke image analysis in real-life scenarios.

Dementia in Convolutional Neural Networks: Using Deep Learning Models to Simulate Neurodegeneration of the Visual System.

Although current research aims to improve deep learning networks by applying knowledge about the healthy human brain and vice versa, the potential of using such networks to model and study neurodegenerative diseases remains largely unexplored. In this work, we present an in-depth feasibility study modeling progressive dementia in silico with deep convolutional neural networks. Therefore, networks were trained to perform visual object recognition and then progressively injured by applying neuronal as well as synaptic injury. After each iteration of injury, network object recognition accuracy, saliency map similarity between the intact and injured networks, and internal activations of the degenerating models were evaluated. The evaluation revealed that cognitive function of the network progressively decreased with increasing injury load whereas this effect was much more pronounced for synaptic damage. The effects of neurodegeneration found for the in silico model are especially similar to the loss of visual cognition seen in patients with posterior cortical atrophy.

The use of acrylic yarn modified with amidoxime and carboxylate-containing polymer for lead removal from drinking water.

Amidoxime and carboxylate-containing polymer adsorbents derived from acrylic yarn exhibit high adsorption capacity for lead(ii) (Pb2+) ions in water. The adsorption process follows pseudo-second-order kinetics and fits the extended Langmuir isotherm model with the maximum adsorption capacity of Pb2+ with 238 mg lead per gram of the fiber at room temperature. Endothermic (ΔH° = 20.3 kJ per mole), spontaneous, and with the increase in the entropy of Pb2+ adsorption was observed from the thermodynamic studies. Dynamic column adsorption experiments showed that the fiber can process 4.3 L of water spiked with 1 ppm of lead(ii) solution at a flow rate of 4.4 mL per min under the specified conditions. The selectivity of Pb2+ with the competitive metal ions showed varying results with highly selective for Pb2+ in a binary solution with sodium and calcium and varying degrees of competitiveness with transition metal ions. This efficient and easily prepared fiber adsorbent appears to be a promising new material for the remediation of lead-contaminated aquatic environments and potable waters.

An analysis of the effects of limited training data in distributed learning scenarios for brain age prediction.

OBJECTIVE: Distributed learning avoids problems associated with central data collection by training models locally at each site. This can be achieved by federated learning (FL) aggregating multiple models that were trained in parallel or training a single model visiting sites sequentially, the traveling model (TM). While both approaches have been applied to medical imaging tasks, their performance in limited local data scenarios remains unknown. In this study, we specifically analyze FL and TM performances when very small sample sizes are available per site. MATERIALS AND METHODS: 2025 T1-weighted magnetic resonance imaging scans were used to investigate the effect of sample sizes on FL and TM for brain age prediction. We evaluated models across 18 scenarios varying the number of samples per site (1, 2, 5, 10, and 20) and the number of training rounds (20, 40, and 200). RESULTS: Our results demonstrate that the TM outperforms FL, for every sample size examined. In the extreme case when each site provided only one sample, FL achieved a mean absolute error (MAE) of 18.9 ± 0.13 years, while the TM achieved a MAE of 6.21 ± 0.50 years, comparable to central learning (MAE = 5.99 years). DISCUSSION: Although FL is more commonly used, our study demonstrates that TM is the best implementation for small sample sizes. CONCLUSION: The TM offers new opportunities to apply machine learning models in rare diseases and pediatric research but also allows even small hospitals to contribute small datasets.

Modeling Neurodegeneration in silico With Deep Learning.

Deep neural networks, inspired by information processing in the brain, can achieve human-like performance for various tasks. However, research efforts to use these networks as models of the brain have primarily focused on modeling healthy brain function so far. In this work, we propose a paradigm for modeling neural diseases in silico with deep learning and demonstrate its use in modeling posterior cortical atrophy (PCA), an atypical form of Alzheimer's disease affecting the visual cortex. We simulated PCA in deep convolutional neural networks (DCNNs) trained for visual object recognition by randomly injuring connections between artificial neurons. Results showed that injured networks progressively lost their object recognition capability. Simulated PCA impacted learned representations hierarchically, as networks lost object-level representations before category-level representations. Incorporating this paradigm in computational neuroscience will be essential for developing in silico models of the brain and neurological diseases. The paradigm can be expanded to incorporate elements of neural plasticity and to other cognitive domains such as motor control, auditory cognition, language processing, and decision making.

An Analysis of the Vulnerability of Two Common Deep Learning-Based Medical Image Segmentation Techniques to Model Inversion Attacks.

Recent research in computer vision has shown that original images used for training of deep learning models can be reconstructed using so-called inversion attacks. However, the feasibility of this attack type has not been investigated for complex 3D medical images. Thus, the aim of this study was to examine the vulnerability of deep learning techniques used in medical imaging to model inversion attacks and investigate multiple quantitative metrics to evaluate the quality of the reconstructed images. For the development and evaluation of model inversion attacks, the public LPBA40 database consisting of 40 brain MRI scans with corresponding segmentations of the gyri and deep grey matter brain structures were used to train two popular deep convolutional neural networks, namely a U-Net and SegNet, and corresponding inversion decoders. Matthews correlation coefficient, the structural similarity index measure (SSIM), and the magnitude of the deformation field resulting from non-linear registration of the original and reconstructed images were used to evaluate the reconstruction accuracy. A comparison of the similarity metrics revealed that the SSIM is best suited to evaluate the reconstruction accuray, followed closely by the magnitude of the deformation field. The quantitative evaluation of the reconstructed images revealed SSIM scores of 0.73±0.12 and 0.61±0.12 for the U-Net and the SegNet, respectively. The qualitative evaluation showed that training images can be reconstructed with some degradation due to blurring but can be correctly matched to the original images in the majority of the cases. In conclusion, the results of this study indicate that it is possible to reconstruct patient data used for training of convolutional neural networks and that the SSIM is a good metric to assess the reconstruction accuracy.

Supervised machine learning tools: a tutorial for clinicians.

In an increasingly data-driven world, artificial intelligence is expected to be a key tool for converting big data into tangible benefits and the healthcare domain is no exception to this. Machine learning aims to identify complex patterns in multi-dimensional data and use these uncovered patterns to classify new unseen cases or make data-driven predictions. In recent years, deep neural networks have shown to be capable of producing results that considerably exceed those of conventional machine learning methods for various classification and regression tasks. In this paper, we provide an accessible tutorial of the most important supervised machine learning concepts and methods, including deep learning, which are potentially the most relevant for the medical domain. We aim to take some of the mystery out of machine learning and depict how machine learning models can be useful for medical applications. Finally, this tutorial provides a few practical suggestions for how to properly design a machine learning model for a generic medical problem.

Building machine learning models without sharing patient data: A simulation-based analysis of distributed learning by ensembling.

The development of machine learning solutions in medicine is often hindered by difficulties associated with sharing patient data. Distributed learning aims to train machine learning models locally without requiring data sharing. However, the utility of distributed learning for rare diseases, with only a few training examples at each contributing local center, has not been investigated. The aim of this work was to simulate distributed learning models by ensembling with artificial neural networks (ANN), support vector machines (SVM), and random forests (RF) and evaluate them using four medical datasets. Distributed learning by ensembling locally trained agents improved performance compared to models trained using the data from a single institution, even in cases where only a very few training examples are available per local center. Distributed learning improved when more locally trained models were added to the ensemble. Local class imbalance reduced distributed SVM performance but did not impact distributed RF and ANN classification. Our results suggest that distributed learning by ensembling can be used to train machine learning models without sharing patient data and is suitable to use with small datasets.

Injectable hydrogel enables local and sustained co-delivery to the brain: Two clinically approved biomolecules, cyclosporine and erythropoietin, accelerate functional recovery in rat model of stroke.

Therapeutic delivery to the brain is limited by the blood-brain barrier and is exacerbated by off-target effects associated with systemic delivery, thereby precluding many potential therapies from even being tested. Given the systemic side effects of cyclosporine and erythropoietin, systemic administration would be precluded in the context of stroke, leaving only the possibility of local delivery. We wondered if direct delivery to the brain would allow new reparative therapeutics, such as these, to be identified for stroke. Using a rodent model of stroke, we employed an injectable drug delivery hydrogel strategy to circumvent the blood-brain barrier and thereby achieved, for the first time, local and sustained co-release to the brain of cyclosporine and erythropoietin. Both drugs diffused to the sub-cortical neural stem and progenitor cell (NSPC) niche and were present in the brain for at least 32 days post-stroke. Each drug had a different outcome on brain tissue: cyclosporine increased plasticity in the striatum while erythropoietin stimulated endogenous NSPCs. Only their co-delivery, but not either drug alone, accelerated functional recovery and improved tissue repair. This platform opens avenues for hitherto untested therapeutic combinations to promote regeneration and repair after stroke.

Local Delivery of Brain-Derived Neurotrophic Factor Enables Behavioral Recovery and Tissue Repair in Stroke-Injured Rats.

IMPACT STATEMENT: We developed a biocomposite that can be mixed with brain-derived neurotrophic factor (BDNF) and dispensed onto the surface of the brain to provide sustained, local release of the protein using a procedure that avoids additional damage to neural tissue. The composite is simple to fabricate, and provides sustained release without nanoparticle encapsulation of BDNF, preserving material and protein bioactivity. We demonstrate that when delivered epicortically to a rat model of stroke, this composite allows BDNF to diffuse into the brain, resulting in enhanced behavioral recovery and synaptic plasticity in the contralesional hemisphere.

Initial cell maturity changes following transplantation in a hyaluronan-based hydrogel and impacts therapeutic success in the stroke-injured rodent brain.

Ischemic stroke results in a loss of neurons for which there are no available clinical strategies to stimulate regeneration. While preclinical studies have demonstrated that functional recovery can be obtained by transplanting an exogenous source of neural progenitors into the brain, it remains unknown at which stage of neuronal maturity cells will provide the most benefit. We investigated the role of neuronal maturity on cell survival, differentiation, and long-term sensorimotor recovery in stroke-injured rats using a population of human cortically-specified neuroepithelial progenitor cells (cNEPs) delivered in a biocompatible, bioresorbable hyaluronan/methylcellulose hydrogel. We demonstrate that transplantation of immature cNEPs result in the greatest tissue and functional repair, relative to transplantation of more mature neurons. The transplantation process itself resulted in the least cell death and phenotypic changes in the immature cNEPs, and the greatest acute cell death in the mature cells. The latter negatively impacted host tissue and negated any potential positive effects associated with cell maturity and the hydrogel vehicle, which itself showed some functional and tissue benefit. Moreover, we show that more mature cell populations are drastically altered during the transplantation process itself. The phenotype of the cells before and after transplantation had an enormous impact on their survival and the consequent tissue and behavioral response, emphasizing the importance of characterizing injected cells in transplantation studies more broadly.

Encapsulation-free controlled release: Electrostatic adsorption eliminates the need for protein encapsulation in PLGA nanoparticles.

Encapsulation of therapeutic molecules within polymer particles is a well-established method for achieving controlled release, yet challenges such as low loading, poor encapsulation efficiency, and loss of protein activity limit clinical translation. Despite this, the paradigm for the use of polymer particles in drug delivery has remained essentially unchanged for several decades. By taking advantage of the adsorption of protein therapeutics to poly(lactic-co-glycolic acid) (PLGA) nanoparticles, we demonstrate controlled release without encapsulation. In fact, we obtain identical, burst-free, extended-release profiles for three different protein therapeutics with and without encapsulation in PLGA nanoparticles embedded within a hydrogel. Using both positively and negatively charged proteins, we show that short-range electrostatic interactions between the proteins and the PLGA nanoparticles are the underlying mechanism for controlled release. Moreover, we demonstrate tunable release by modifying nanoparticle concentration, nanoparticle size, or environmental pH. These new insights obviate the need for encapsulation and offer promising, translatable strategies for a more effective delivery of therapeutic biomolecules.

Circumventing the blood-brain barrier: Local delivery of cyclosporin A stimulates stem cells in stroke-injured rat brain.

Drug delivery to the central nervous system is limited by the blood-brain barrier, which can be circumvented by local delivery. In applications of stroke therapy, for example, stimulation of endogenous neural stem/progenitor cells (NSPCs) by cyclosporin A (CsA) is promising. However, current strategies rely on high systemic drug doses to achieve small amounts of CsA in the brain tissue, resulting in systemic toxicity and undesirable global immunosuppression. Herein we describe the efficacy of local CsA delivery to the stroke-injured rat brain using an epi-cortically injected hydrogel composed of hyaluronan and methylcellulose (HAMC). CsA was encapsulated in poly(lactic-co-glycolic acid) microparticles dispersed in HAMC, allowing for its sustained release over 14days in vivo. Tissue penetration was sufficient to provide sustained CsA delivery to the sub-cortical NSPC niche. In comparison to systemic delivery using an osmotic minipump, HAMC achieved higher CsA concentrations in the brain while significantly reducing drug exposure in other organs. HAMC alone was beneficial in the stroke-injured rat brain, significantly reducing the stroke infarct volume relative to untreated stroke-injured controls. The combination of HAMC and local CsA release increased the number of proliferating cells in the lateral ventricles - the NSPC niche in the adult brain. Thus, we demonstrate a superior method of drug delivery to the rat brain that provides dual benefits of tissue protection and endogenous NSPC stimulation after stroke.

A hydrogel composite system for sustained epi-cortical delivery of Cyclosporin A to the brain for treatment of stroke.

Stimulation of endogenous neural stem/progenitor cells (NSPCs) with therapeutic factors holds potential for the treatment of stroke. Cyclosporin A (CsA) is a particularly promising candidate molecule because it has been shown to act as a survival factor for these cells over a period of weeks both in vitro and in vivo; however, systemically-delivered CsA compromises the entire immune system, necessitating sustained localized delivery. Herein we describe a local delivery strategy for CsA using an epi-cortical hydrogel of hyaluronan-methylcellulose (HAMC) as the drug reservoir. Three methods of incorporating the drug into the hydrogel (solubilized, particulate, and poly(lactic-co-glycolic) acid (PLGA) microsphere-encapsulated) resulted in tunable release, spanning a period of hours to weeks. Importantly, PLGA-encapsulated CsA released from the hydrogel had equivalent bioactivity to fresh drug as measured by the neurosphere assay. Moreover, when CsA was released from the PLGA/HAMC composite that was injected on the cortex of adult mice, CsA was detected in the NSPC niche at a constant concentration for at least 24days post-implant. Thus this hydrogel composite system may be promising for the treatment of stroke.

Co-expression vs. co-infection using baculovirus expression vectors in insect cell culture: Benefits and drawbacks.

The baculovirus expression vector system (BEVS) is a versatile and powerful platform for protein expression in insect cells. With the ability to approach similar post-translational modifications as in mammalian cells, the BEVS offers a number of advantages including high levels of expression as well as an inherent safety during manufacture and of the final product. Many BEVS products include proteins and protein complexes that require expression from more than one gene. This review examines the expression strategies that have been used to this end and focuses on the distinguishing features between those that make use of single polycistronic baculovirus (co-expression) and those that use multiple monocistronic baculoviruses (co-infection). Three major areas in which researchers have been able to take advantage of co-expression/co-infection are addressed, including compound structure-function studies, insect cell functionality augmentation, and VLP production. The core of the review discusses the parameters of interest for co-infection and co-expression with time of infection (TOI) and multiplicity of infection (MOI) highlighted for the former and the choice of promoter for the latter. In addition, an overview of modeling approaches is presented, with a suggested trajectory for future exploration. The review concludes with an examination of the gaps that still remain in co-expression/co-infection knowledge and practice.

Estimation of mental effort in learning visual search by measuring pupil response.

Perceptual learning refers to the improvement of perceptual sensitivity and performance with training. In this study, we examined whether learning is accompanied by a release from mental effort on the task, leading to automatization of the learned task. For this purpose, we had subjects conduct a visual search for a target, defined by a combination of orientation and spatial frequency, while we monitored their pupil size. It is well known that pupil size reflects the strength of mental effort invested in a task. We found that pupil size increased rapidly as the learning proceeded in the early phase of training and decreased at the later phase to a level half of its maximum value. This result does not support the simple automatization hypothesis. Instead, it suggests that the mental effort and behavioral performance reflect different aspects of perceptual learning. Further, mental effort would be continued to be invested to maintain good performance at a later stage of training.

The effect of retinal illuminance on visual motion priming.

The perceived direction of a directionally ambiguous stimulus is influenced by the moving direction of a preceding priming stimulus. Previous studies have shown that a brief priming stimulus induces positive motion priming, in which a subsequent directionally ambiguous stimulus is perceived to move in the same direction as the primer, while a longer priming stimulus induces negative priming, in which the following ambiguous stimulus is perceived to move in the opposite direction of the primer. The purpose of this study was to elucidate the underlying mechanism of motion priming by examining how retinal illuminance and velocity of the primer influences the perception of priming. Subjects judged the perceived direction of 180-deg phase-shifted (thus directionally ambiguous) sine-wave gratings displayed immediately after the offset of a primer stimulus. We found that perception of motion priming was greatly modulated by the retinal illuminance and velocity of the primer. Under low retinal illuminance, positive priming nearly disappeared even when the effective luminance contrast was equated between different conditions. Positive priming was prominent when the velocity of the primer was low, while only negative priming was observed when the velocity was high. These results suggest that the positive motion priming is induced by a higher-order mechanism that tracks prominent features of the visual stimulus, while a directionally selective motion mechanism induces negative motion priming.