Deep learning in MRI-guided radiation therapy: A systematic review.

Recent advances in MRI-guided radiation therapy (MRgRT) and deep learning techniques encourage fully adaptive radiation therapy (ART), real-time MRI monitoring, and the MRI-only treatment planning workflow. Given the rapid growth and emergence of new state-of-the-art methods in these fields, we systematically review 197 studies written on or before December 31, 2022, and categorize the studies into the areas of image segmentation, image synthesis, radiomics, and real time MRI. Building from the underlying deep learning methods, we discuss their clinical importance and current challenges in facilitating small tumor segmentation, accurate x-ray attenuation information from MRI, tumor characterization and prognosis, and tumor motion tracking. In particular, we highlight the recent trends in deep learning such as the emergence of multi-modal, visual transformer, and diffusion models.

Measuring red blood cell shape in the human retina.

The free diameter of a red blood cell generally exceeds the lumen diameter of capillaries in the central nervous system, requiring significant cellular deformation. However, the deformations undertaken are not well established under natural conditions due to the difficulty in observing corpuscular flow in vivo. Here we describe a novel, to the best of our knowledge, method to noninvasively study the shape of red blood cells as they traverse the narrow capillary networks of the living human retina, using high-speed adaptive optics. One hundred and twenty-three capillary vessels were analyzed in three healthy subjects. For each capillary, image data were motion-compensated and then averaged over time to reveal the appearance of the blood column. Data from hundreds of red blood cells were used to profile the average cell in each vessel. Diverse cellular geometries were observed across lumens ranging from 3.2 to 8.4 µm in diameter. As capillaries narrowed, cells transitioned from rounder to more elongated shapes and from being counter-aligned to aligned with the axis of flow. Remarkably, in many vessels the red blood cells maintained an oblique orientation relative to the axis of flow.

Structural basis for proteolytic processing of Aspergillus sojae α-glucosidase L with strong transglucosylation activity.

An α-glucosidase from Aspergillus sojae, AsojAgdL, exhibits strong transglucosylation activity to produce α-1,6-glucosidic linkages. The most remarkable structural feature of AsojAgdL is that residues 457-560 of AsojAgdL (designated the NC sequence) is not conserved in other glycoside hydrolase family 31 enzymes, and part of this NC sequence is proteolytically cleaved during its maturation. In this study, the enzyme was expressed in Pichia pastoris, and electrophoretic analysis indicated that the recombinant enzyme, rAsojAgdL, consisted of two polypeptide chains, as observed in the case of the enzyme produced in an Aspergillus strain. The crystal structure of rAsojAgdL was determined in complex with the substrate analog trehalose. Electron density corresponding to residues 496-515 of the NC sequence was not seen, and there were no α-helices or ß-strands except for a short α-helix in the structures of residues 457-495 and residues 516-560, both of which belong to the NC sequence. The residues 457-495 and the residues 516-560 both formed extra components of the catalytic domain. The residues 457-495 constituted the entrance of the catalytic pocket of rAsojAgdL, and Gly467, Asp468, Pro469, and Pro470 in the NC sequence were located within 4 Å of Trp400, a key residue involved in binding of the substrate. The results suggest that the proteolytic processing of the NC sequence is related to the formation of the catalytic pocket of AsojAgdL.

In Situ Growth of Crystalline and Polymer-Incorporated Amorphous ZIFs in Polybenzimidazole Achieving Hierarchical Nanostructures for Carbon Capture.

Mixed matrix materials (MMMs) hold great potential for membrane gas separations by merging nanofillers with unique nanostructures and polymers with excellent processability. In situ growth of the nanofillers is adapted to mitigate interfacial incompatibility to avoid the selectivity loss. Surprisingly, functional polymers have not been exploited to co-grow the nanofillers for membrane applications. Herein, in situ synergistic growth of crystalline zeolite imidazole framework-8 (ZIF-8) in polybenzimidazole (PBI), creating highly porous structures with high gas permeability, is demonstrated. More importantly, PBI contains benzimidazole groups (similar to the precursor for ZIF-8, i.e., 2-methylimidazole) and induces the formation of amorphous ZIFs, enhancing interfacial compatibility and creating highly size-discriminating bottlenecks. For instance, the formation of 15 mass% ZIF-8 in PBI improves H2 permeability and H2 /CO2 selectivity by ≈100% at 35 °C, breaking the permeability/selectivity tradeoff. This work unveils a new platform of MMMs comprising functional polymer-incorporated amorphous ZIFs with hierarchical nanostructures for various applications.

A wide-field micro-computed tomography detector: micron resolution at half-centimetre scale.

Ideal three-dimensional imaging of complex samples made up of micron-scale structures extending over mm to cm, such as biological tissues, requires both wide field of view and high resolution. For existing optics and detectors used for micro-CT (computed tomography) imaging, sub-micron pixel resolution can only be achieved for fields of view of <2 mm. This article presents a unique detector system with a 6 mm field-of-view image circle and 0.5 µm pixel size that can be used in micro-CT units utilizing both synchrotron and commercial X-ray sources. A resolution-test pattern with linear microstructures and whole adult Daphnia magna were imaged at beamline 8.3.2 of the Berkeley Advanced Light Source. Volumes of 10000 × 10000 × 7096 isotropic 0.5 µm voxels were reconstructed over a 5.0 mm × 3.5 mm field of view. Measurements in the projection domain confirmed a 0.90 µm measured spatial resolution that is largely Nyquist-limited. This unprecedented combination of field of view and resolution dramatically reduces the need for sectional scans and computational stitching for large samples, ultimately offering the means to elucidate changes in tissue and cellular morphology in the context of larger, whole, intact model organisms and specimens. This system is also anticipated to benefit micro-CT imaging in materials science, microelectronics, agricultural science and biomedical engineering.

Nickel-Based Membrane Electrodes Enable High-Rate Electrochemical Ammonia Recovery.

Wastewater contains significant amounts of nitrogen that can be recovered and valorized as fertilizers and chemicals. This study presents a new membrane electrode coupled with microbial electrolysis that demonstrates very efficient ammonia recovery from synthetic centrate. The process utilizes the electrical potential across electrodes to drive NH4+ ions toward the hydrophilic nickel top layer on a gas-stripping membrane cathode, which takes advantage of surface pH increase to realize spontaneous NH3 production and separation. Compared with a control configuration with conventionally separated electrode and hydrophobic membrane, the integrated membrane electrode showed 40% higher NH3-N recovery rate (36.2 ± 1.2 gNH3-N/m2/d) and 11% higher current density. The energy consumption was 1.61 ± 0.03 kWh/kgNH3-N, which was 20% lower than the control and 70-90% more efficient than competing electrochemical nitrogen recovery processes (5-12 kWh/kgNH3-N). Besides, the negative potential on membrane electrode repelled negatively charged organics and microbes thus reduced fouling. In addition to describing the system's performance, we explored the underlying mechanisms governing the reactions, which confirmed the viability of this process for efficient wastewater-ammonia recovery. Furthermore, the nickel-based membrane electrode showed excellent water entry pressure (∼41 kPa) without leakage, which was much higher than that of PTFE/PDMS-based cathodes (∼1.8 kPa). The membrane electrode also showed superb flexibility (180° bend) and can be easily fabricated at low cost (<20 $/m2).

Influence of support-layer deformation on the intrinsic resistance of thin film composite membranes.

It is commonly believed that the overall permeation resistance of thin film composite (TFC) membranes is dictated by the crosslinked, ultrathin polyamide barrier layer, while the porous support merely serves as the mechanical support. Although this assumption might be the case under low transmembrane pressure, it becomes questionable under high transmembrane pressure. A highly porous support normally yields under a pressure of a few MPa, which can result in a significant level of compressive strain that may significantly increase the resistance to permeation. However, quantifying the influence of porous support deformation on the overall resistance of the TFC membrane is challenging. In particular, it is difficult to determine the deformation/strain of the membrane during active separation. In this study, we use nanoimprint lithography (NIL) to achieve precise compressive deformation in commercial TFC membranes. By adjusting the NIL conditions, membranes were compressed to strain levels up to 60%. SEM and AFM measurements showed that the compression had minimal impact on the barrier-layer surface morphology and total surface area with most of the deformation occurring in the support layer. DI water permeation measurements revealed that the water flux reduction decreases with an increase of strain level. Most significantly, the intrinsic membrane resistance showed negligible changes at strain levels lower than 30%-40%, but increased exponentially at higher strain levels, reaching 250%-500% of pristine (unstrained) membrane values. Using a resistance-in-series model, the strain dependency of the TFC membrane resistance can be described.

Implementation of two distinct wavelengths to induce multistage polymerization in shape memory materials and nanoimprint lithography.

Here, a process is introduced for forming dual stage thiol-Michael/acrylate hybrid networks photocured by two different wavelengths, demonstrating its use in nanoimprint lithography (NIL) and shape memory materials. Initiated with a visible light sensitive photobase and a UV-sensitive radical initiator, thiol-Michael-acrylate hybrid polymerizations were programmed to proceed sequentially and orthogonally, with base-catalyzed thiol-Michael photopolymerization as the first stage and radical mediated acrylate photopolymerization as the second stage. By regulating the photopolymerization formulations, i.e. thiol-to-acrylate ratios, initiator loadings and irradiation conditions, a series of materials with highly tunable mechanical performance was achieved, with ultimate Tg values ranging from 23 to 70 °C. With a photopatternable first stage and a readily reconfigurable second stage, its implementation in nanoimprint lithography (NIL) enabled surface features on the scale of 10 nm to be formed on a photopatterned substrate. Additionally, the dual stage polymer results in a relatively homogenous polymer network with a narrow glass transition temperature (Tg), which enables rapid response in applications as shape memory materials, with shape-fixity values above 95% and shaperecovery values above 99%. With its unique photocuring process and programmable mechanical properties, the two color light controlled photopolymerization can be exploited as a useful tool in a wide range of materials science applications.

Influences of Substrate Adhesion and Particle Size on the Shape Memory Effect of Polystyrene Particles.

Formulations and applications of micro- and nanoscale polymer particles have proliferated rapidly in recent years, yet knowledge of their mechanical behavior has not grown accordingly. In this study, we examine the ways that compressive strain, substrate surface energy, and particle size influence the shape memory cycle of polystyrene particles. Using nanoimprint lithography, differently sized particles are programmed into highly deformed, temporary shapes in contact with substrates of differing surface energies. Atomic force microscopy is used to obtain in situ measurements of particle shape recovery kinetics, and scanning electron microscopy is employed to assess differences in the profiles of particles at the conclusion of the shape memory cycle. Finally, finite element models are used to investigate the growing impact of surface energies at smaller length scales. Results reveal that the influence of substrate adhesion on particle recovery is size-dependent and can become dominating at submicron length scales.

Capillary rupture of suspended polymer concentric rings.

We present the first experimental study on the simultaneous capillary instability amongst viscous concentric rings suspended atop an immiscible medium. The rings ruptured upon annealing, with three types of phase correlation between neighboring rings. In the case of weak substrate confinement, the rings ruptured independently when they were sparsely distanced, but via an out-of-phase mode when packed closer. If the substrate confinement was strong, the rings would rupture via an in-phase mode, resulting in radially aligned droplets. The concentric ring geometry caused a competition between the phase correlation of neighboring rings and the kinetically favorable wavelength, yielding an intriguing, recursive surface pattern. This frustrated pattern formation behavior was accounted for by a scaling analysis.

Changes to the shape, orientation and packing of red cells as a function of retinal capillary size.

The free diameter of a red blood cell exceeds the lumen diameter of capillaries in the central nervous system, requiring significant deformation of cells. However the deformations undertaken in vivo are not well established due to the difficulty in observing cellular capillary flow in living human tissue. Here, we used high resolution adaptive optics imaging to non-invasively track 17,842 red blood cells in transit through 121 unique capillary segments of diameter 8 µm or less in the retina of 3 healthy human subjects. Within each vessel, a 2D en face profile was generated for the "average cell", whose shape was then inferred in 3D based on the key assumption of a circular capillary cross-section. From this we estimated the average volume, surface area, orientation, and separation between red cells within each capillary tube. Our results showed a network filtration effect, whereby narrower vessels were more likely to contain smaller cells (defined by surface area, which is thought not to vary during a cell's passage through the vascular system). A bivariate linear model showed that for larger cells in narrower vessels: cells re-orient themselves to align with the flow axis, their shape becomes more elongated, there are longer gaps between successive cells, and remarkably, that cell volume is less which implies the ejection of water from cells to facilitate capillary transit. Taken together, these findings suggest that red cells pass through retinal capillaries with some reluctance. A biphasic distribution for cell orientation and separation was evident, indicating a "tipping point" for vessels narrower than approx. 5 µm. This corresponds closely to the typical capillary lumen diameter, and may maximize sensitivity of cellular flow to small changes in diameter. We suggest that the minimization of unnecessary oxygen exchange, and hence of damage via reactive oxygen pathways, may have provided evolutionary pressure to ensure that capillary lumens are generally narrower than red blood cells.

Hierarchically porous and single Zn atom-embedded carbon molecular sieves for H2 separations.

Hierarchically porous materials containing sub-nm ultramicropores with molecular sieving abilities and microcavities with high gas diffusivity may realize energy-efficient membranes for gas separations. However, rationally designing and constructing such pores into large-area membranes enabling efficient H2 separations remains challenging. Here, we report the synthesis and utilization of hybrid carbon molecular sieve membranes with well-controlled nano- and micro-pores and single zinc atoms and clusters well-dispersed inside the nanopores via the carbonization of supramolecular mixed matrix materials containing amorphous and crystalline zeolitic imidazolate frameworks. Carbonization temperature is used to fine-tune pore sizes, achieving ultrahigh selectivity for H2/CO2 (130), H2/CH4 (2900), H2/N2 (880), and H2/C2H6 (7900) with stability against water vapor and physical aging during a continuous 120-h test.

Influence of substrate confinement on the phase-correlation in the capillary breakup of arrays of patterned polymer stripes.

We investigated the influence of substrate confinement on the capillary breakup of parallel nonaxisymmetric polymer stripes suspended on top of, or confined between, another immiscible polymer pattern. When the residual layer thickness of the pattern was reasonably large, the PS (or PMMA) stripes confined within PMMA (or PS) trenches broke up, either nucleated, out-of-phase, or without clear phase correlation depending on the geometry and viscosity ratio between the two polymers. In stark contrast, for the two extreme cases of viscosity ratios we studied, in-phase breakup of confined polymer stripes was always observed when the alternating PS/PMMA stripes were formed, that is, without residual layer, regardless of the specific geometry.

Diverse Sample Generation: Pushing the Limit of Generative Data-Free Quantization.

Generative data-free quantization emerges as a practical compression approach that quantizes deep neural networks to low bit-width without accessing the real data. This approach generates data utilizing batch normalization (BN) statistics of the full-precision networks to quantize the networks. However, it always faces the serious challenges of accuracy degradation in practice. We first give a theoretical analysis that the diversity of synthetic samples is crucial for the data-free quantization, while in existing approaches, the synthetic data completely constrained by BN statistics experimentally exhibit severe homogenization at distribution and sample levels. This paper presents a generic Diverse Sample Generation (DSG) scheme for the generative data-free quantization, to mitigate detrimental homogenization. We first slack the statistics alignment for features in the BN layer to relax the distribution constraint. Then, we strengthen the loss impact of the specific BN layers for different samples and inhibit the correlation among samples in the generation process, to diversify samples from the statistical and spatial perspectives, respectively. Comprehensive experiments show that for large-scale image classification tasks, our DSG can consistently quantization performance on different neural architectures, especially under ultra-low bit-width. And data diversification caused by our DSG brings a general gain to various quantization-aware training and post-training quantization approaches, demonstrating its generality and effectiveness.

SREBP Regulation of Lipid Metabolism in Liver Disease, and Therapeutic Strategies.

Sterol regulatory element-binding proteins (SREBPs) are master transcription factors that play a crucial role in regulating genes involved in the biogenesis of cholesterol, fatty acids, and triglycerides. As such, they are implicated in several serious liver diseases, including non-alcoholic fatty liver disease (NAFLD), non-alcoholic steatohepatitis (NASH), fibrosis, and hepatocellular carcinoma (HCC). SREBPs are subject to regulation by multiple cofactors and critical signaling pathways, making them an important target for therapeutic interventions. In this review, we first introduce the structure and activation of SREBPs, before focusing on their function in liver disease. We examine the mechanisms by which SREBPs regulate lipogenesis, explore how alterations in these processes are associated with liver disease, and evaluate potential therapeutic strategies using small molecules, natural products, or herb extracts that target these pathways. Through this analysis, we provide new insights into the versatility and multitargets of SREBPs as factors in the modulation of different physiological stages of liver disease, highlighting their potential targets for therapeutic treatment.

Hippocampus Substructure Segmentation Using Morphological Vision Transformer Learning.

Background: The hippocampus plays a crucial role in memory and cognition. Because of the associated toxicity from whole brain radiotherapy, more advanced treatment planning techniques prioritize hippocampal avoidance, which depends on an accurate segmentation of the small and complexly shaped hippocampus. Purpose: To achieve accurate segmentation of the anterior and posterior regions of the hippocampus from T1 weighted (T1w) MRI images, we developed a novel model, Hippo-Net, which uses a mutually enhanced strategy. Methods: The proposed model consists of two major parts: 1) a localization model is used to detect the volume-of-interest (VOI) of hippocampus. 2) An end-to-end morphological vision transformer network is used to perform substructures segmentation within the hippocampus VOI. The substructures include the anterior and posterior regions of the hippocampus, which are defined as the hippocampus proper and parts of the subiculum. The vision transformer incorporates the dominant features extracted from MRI images, which are further improved by learning-based morphological operators. The integration of these morphological operators into the vision transformer increases the accuracy and ability to separate hippocampus structure into its two distinct substructures.A total of 260 T1w MRI datasets from Medical Segmentation Decathlon dataset were used in this study. We conducted a five-fold cross-validation on the first 200 T1w MR images and then performed a hold-out test on the remaining 60 T1w MR images with the model trained on the first 200 images. The segmentations were evaluated with two indicators, 1) multiple metrics including the Dice similarity coefficient (DSC), 95th percentile Hausdorff distance (HD95), mean surface distance (MSD), volume difference (VD) and center-of-mass distance (COMD); 2) Volumetric Pearson correlation analysis. Results: In five-fold cross-validation, the DSCs were 0.900±0.029 and 0.886±0.031 for the hippocampus proper and parts of the subiculum, respectively. The MSD were 0.426±0.115mm and 0.401±0.100 mm for the hippocampus proper and parts of the subiculum, respectively. Conclusions: The proposed method showed great promise in automatically delineating hippocampus substructures on T1w MRI images. It may facilitate the current clinical workflow and reduce the physicians' effort.

Deep Learning in MRI-guided Radiation Therapy: A Systematic Review.

MRI-guided radiation therapy (MRgRT) offers a precise and adaptive approach to treatment planning. Deep learning applications which augment the capabilities of MRgRT are systematically reviewed. MRI-guided radiation therapy offers a precise, adaptive approach to treatment planning. Deep learning applications which augment the capabilities of MRgRT are systematically reviewed with emphasis placed on underlying methods. Studies are further categorized into the areas of segmentation, synthesis, radiomics, and real time MRI. Finally, clinical implications, current challenges, and future directions are discussed.

BiFSMNv2: Pushing Binary Neural Networks for Keyword Spotting to Real-Network Performance.

Deep neural networks, such as the deep-FSMN, have been widely studied for keyword spotting (KWS) applications while suffering expensive computation and storage. Therefore, network compression technologies such as binarization are studied to deploy KWS models on edge. In this article, we present a strong yet efficient binary neural network for KWS, namely, BiFSMNv2, pushing it to the real-network accuracy performance. First, we present a dual-scale thinnable 1-bit-architecture (DTA) to recover the representation capability of the binarized computation units by dual-scale activation binarization and liberate the speedup potential from an overall architecture perspective. Second, we also construct a frequency-independent distillation (FID) scheme for KWS binarization-aware training, which distills the high-and low-frequency components independently to mitigate the information mismatch between full-precision and binarized representations. Moreover, we propose the learning propagation binarizer (LPB), a general and efficient binarizer that enables the forward and backward propagation of binary KWS networks to be continuously improved through learning. We implement and deploy BiFSMNv2 on ARMv8 real-world hardware with a novel fast bitwise computation kernel (FBCK), which is proposed to fully use registers and increase instruction throughput. Comprehensive experiments show our BiFSMNv2 outperforms the existing binary networks for KWS by convincing margins across different datasets and achieves comparable accuracy with the full-precision networks (only a tiny 1.51% drop on Speech Commands V1-12). We highlight that benefiting from the compact architecture and optimized hardware kernel, BiFSMNv2 can achieve an impressive 25.1 × speedup and 20.2 × storage-saving on edge hardware.

Hippocampus substructure segmentation using morphological vision transformer learning.

The hippocampus plays a crucial role in memory and cognition. Because of the associated toxicity from whole brain radiotherapy, more advanced treatment planning techniques prioritize hippocampal avoidance, which depends on an accurate segmentation of the small and complexly shaped hippocampus. To achieve accurate segmentation of the anterior and posterior regions of the hippocampus from T1 weighted (T1w) MR images, we developed a novel model, Hippo-Net, which uses a cascaded model strategy. The proposed model consists of two major parts: (1) a localization model is used to detect the volume-of-interest (VOI) of hippocampus. (2) An end-to-end morphological vision transformer network (Franchietal2020Pattern Recognit.102107246, Ranemetal2022 IEEE/CVF Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW) pp 3710-3719) is used to perform substructures segmentation within the hippocampus VOI. The substructures include the anterior and posterior regions of the hippocampus, which are defined as the hippocampus proper and parts of the subiculum. The vision transformer incorporates the dominant features extracted from MR images, which are further improved by learning-based morphological operators. The integration of these morphological operators into the vision transformer increases the accuracy and ability to separate hippocampus structure into its two distinct substructures. A total of 260 T1w MRI datasets from medical segmentation decathlon dataset were used in this study. We conducted a five-fold cross-validation on the first 200 T1w MR images and then performed a hold-out test on the remaining 60 T1w MR images with the model trained on the first 200 images. In five-fold cross-validation, the Dice similarity coefficients were 0.900 ± 0.029 and 0.886 ± 0.031 for the hippocampus proper and parts of the subiculum, respectively. The mean surface distances (MSDs) were 0.426 ± 0.115 mm and 0.401 ± 0.100 mm for the hippocampus proper and parts of the subiculum, respectively. The proposed method showed great promise in automatically delineating hippocampus substructures on T1w MR images. It may facilitate the current clinical workflow and reduce the physicians' effort.

Palladium-Percolated Networks Enabled by Low Loadings of Branched Nanorods for Enhanced H2 Separations.

Nanoparticles (NPs) at high loadings are often used in mixed matrix membranes (MMMs) to improve gas separation properties, but they can lead to defects and poor processability that impede membrane fabrication. Herein, it is demonstrated that branched nanorods (NRs) with controlled aspect ratios can significantly reduce the required loading to achieve superior gas separation properties while maintaining excellent processability, as demonstrated by the dispersion of palladium (Pd) NRs in polybenzimidazole for H2 /CO2 separation. Increasing the aspect ratio from 1 for NPs to 40 for NRs decreases the percolation threshold volume fraction by a factor of 30, from 0.35 to 0.011. An MMM with percolated networks formed by Pd NRs at a volume fraction of 0.039 exhibits H2 permeability of 110 Barrer and H2 /CO2 selectivity of 31 when challenged with simulated syngas at 200 °C, surpassing Robeson's upper bound. This work highlights the advantage of NRs over NPs and nanowires and shows that right-sizing nanofillers in MMMs is critical to construct highly sieving pathways at minimal loadings. This work paves the way for this general feature to be applied across materials systems for a variety of chemical separations.