The University of Sydney

Roof plate secretion of bone morphogenetic proteins (BMPs) directs the cellular fate of sensory neurons during spinal cord development, including the formation of the ascending sensory columns, though their biology is not well understood. Type-II BMP receptor (BMPRII), the cognate receptor, is expressed by neural precursor cells (NPCs) during embryogenesis; however, an in vitro method of enriching BMPRII+ NPCs from fetal spinal cord is absent. Immunofluorescence was undertaken on intact second-trimester human fetal spinal cord using antibodies to BMPRII and Leukemia Inhibitory Factor (LIF). Regions of highest BMPRII+ immunofluorescence localized to sensory columns. Parenchymal and meningeal-associated BMPRII+ vascular cells were identified in both intact fetal spinal cord and cortex by co-positivity with vascular lineage markers, CD34/CD39. LIF immunostaining identified a population of somas concentrated in dorsal and ventral horn interneurons, mirroring the expression of LIF Receptor/CD118. A combination of LIF supplementation and high-density culture maintained culture growth beyond 10 passages, while synergistically increasing the proportion of neurospheres with a stratified cytoarchitecture. These neurospheres were characterized by BMPRII+/MAP2ab+/–/βIII-tubulin+/nestin–/vimentin–/GFAP–/NeuN– surface NPCs surrounding a heterogeneous core of βIII-tubulin+/nestin+/vimentin+/GFAP+/MAP2ab–/NeuN– multipotent precursors. Dissociated cultures from tripotential neurospheres contained neuronal (βIII-tubulin+), astrocytic (GFAP+), and oligodendrocytic (O4+) lineage cells. Fluorescence-activated cell sorting (FACS)-sorted BMPRII+ NPCs were MAP2ab+/–/βIII-tubulin+/GFAP–/O4– in culture. This is the first isolation of BMPRII+ NPCs identified and characterized in human fetal spinal cords. Our data show that LIF combines synergistically with high-density reaggregate cultures to support the organotypic reorganization of neurospheres, characterized by surface BMPRII+ NPCs. Our study has provided a new methodology for an in vitro model capable of amplifying human fetal spinal cord cell numbers for >10 passages. Investigations of the role BMPRII plays in spinal cord development have primarily relied upon mouse and rat models, with interpolations to human development being derived through inference. Because of significant species differences between murine biology and human, including anatomical dissimilarities in central nervous system (CNS) structure, the findings made in murine models cannot be presumed to apply to human spinal cord development and could explain the earlier failure of clinical trials of neural regeneration. For these reasons, our novel human in vitro model offers a tool to better understand neurodevelopmental pathways, including BMP signaling, as well as spinal cord injury research and testing drug therapies. The availability of this in vitro model is especially relevant since pathology is frequently the re-activation or exaggeration of normal developmental processes.

We establish a recursive representation that fully decouples jumps from a large class of multivariate inhomogeneous stochastic differential equations with jumps of general time-state dependent unbounded intensity, not of Lévy-driven type that essentially benefits a lot from independent and stationary increments. The recursive representation, along with a few related ones, are derived by making use of a jump time of the underlying dynamics as an information relay point in passing the past on to a previous iteration step to fill in the missing information on the unobserved trajectory ahead. We prove that the proposed recursive representations are convergent exponentially fast in the limit, and can be represented in a similar form to Picard iterates under the probability measure with its jump component suppressed. On the basis of each iterate, we construct upper and lower bounding functions that are also convergent towards the true solution as the iterations proceed. We provide numerical results to justify our theoretical findings.

The knowledge-based parametric design process of the rotor topology of the synchronous reluctance motor (SynRM) requires a tedious parameter definition, selection, and searching process until an optimal design can be obtained. This research proposes a topology optimization approach based on solid isotropic material with penalization method for the SynRM rotor topology design. The aim is to pursue a more intelligent and automatic design process. Specifically, the proposed method aims to generate the SynRM rotor design of high output torque profile. Meanwhile, the mechanical constraints including structural compliance and stress etc. are also considered for promising the structural reliability of the rotor. Considering the high dimensional constraints caused by the multi-physics performances, the augmented Lagrangian method based optimization framework is developed to solve the minimization problem in a compact scheme. To prove the effectiveness of the developed method, simulations and analysis on the topology optimization of SynRMs are performed in this research. Prototyping and experiment are also conducted for verifying the performance of the optimized structure.

Under the conventional pulse width modulation (PWM) + sinusoidal pulse width modulation (SPWM) hybrid control, the discontinuous conduction mode (DCM) single-leg-integrated boost (SLIB) inverter confronts an over-modulation problem under a wide change in load, leading to the distorted output waveform. One approach to address this issue is to keep the boost-leg duty cycle constant and regulate load power. However, it is not suitable for standalone or PV grid-connected power systems. Therefore, a pulse frequency modulation (PFM) + SPWM hybrid control scheme is proposed to overcome the shortcomings in this paper. Under the proposed scheme, the boost-leg of the inverter works in PFM while the other leg works in SPWM. Since the DC-link voltage is controlled by frequency, instead of load power or duty cycle, the inverter can avoid over-modulation for a wide load range and be applied on occasions where the load power is non-adjustable, which is more practical in comparison with the counterpart. The feasibility of the proposed hybrid control is verified by experiment through a 500 W prototype.

The combination of optical microresonators and the emerging microwave photonic (MWP) sensing has recently drawn great attention, whereas its multi-parameter sensing capability mainly relies on adopting multiple resonance modes. By incorporating deep learning (DL) into MWP sensing, we propose a new sensing paradigm, which has the simplified design, reduced fabrication requirement, and the capability of sensing more than one parameter. The MWP interrogation transforms the spectral response of a single optical resonance (SOR) that can be at arbitrary coupling conditions into the variations of the zero-transmission profile of microwave signals, providing improved interrogation resolution regardless of the resonance parameters. A DL unit is used to exploit the raw interrogation output to simultaneously estimate the target measurands. As the proof-of-concept demonstration, simultaneous temperature and humidity sensing using a SOR is conducted, where the convolutional neural tangent kernel (CNTK) is used as the DL model to reduce the demand for experimental data. The established CNTK-DL model consistently outperforms the support vector regression model that relies on handcrafted features and demonstrates an over 2-fold higher estimation accuracy with the laser drift interference and a lower mean absolute error in the presence of strong noise, showing the power of DL for boosting MWP sensing.

The CAMECA Invizo 6000 atom probe microscope uses ion optics that differ significantly from the local electrode atom probe (LEAP). It uses dual antiparallel deep ultraviolet lasers, a flat counter electrode, and a series of accelerating and decelerating lenses to increase the field-of-view of the specimen without reducing the mass resolving power. In this work we characterise the performance of the Invizo 6000 using three material case studies: a model Al-Mg-Si alloy, a commercially-available Ni-based superalloy, and a Zr alloy, using a combination of air and vacuum-transfer between instruments. The ion optics of the Invizo 6000 significantly increase the field-of-view compared to the same specimen on a LEAP 4000 X Si. We also observe a significant increase in specimen yield, especially for the Zr alloy. These results combine to make the Invizo 6000 well-suited to research projects requiring large analysis volumes, particularly so for traditionally difficult samples such as oxides.

Understanding the occurrence modes of mercury in coal is important as its release poses long-term adverse effects on the environment and human health during coal production and utilization. However, the matter still remains a subject of controversy due to differing results from direct and indirect analyses, which suggest various possible modes of occurrence for mercury in coal. Additionally, the experimental measurement of Hg concentration presents challenges, further contributing to the complexity of the issue. A comprehensive investigation of experiments and molecular simulations is conducted herein. Electron probe microanalysis and elemental mapping analysis show that elemental Hg is concentrated in framboidal pyrites while absent in organic matter. To understand the occurrence modes of mercury in inorganic and organic materials at the atomic level, molecular simulations are performed for Hg2+ adsorption and retention in MMT, pyrite, and kerogen slit nanopores. It is found that the inorganic MMT and pyrite surfaces have a greater adsorption capacity than the organic kerogen surface (pyrite > MMT > kerogen). The outer-sphere adsorption is mainly observed with at least one monolayer of water molecules exiting between the ion and mineral surfaces. MMT has the highest retention for Hg2+ transport as the self-diffusion coefficient is the smallest among the three slit pores (MMT < pyrite < kerogen). The high adsorption and retention originate from the strong Hg2+-mineral interaction. These results suggest that mercury in coal is most likely associated with inorganic minerals instead of organic matter.

New rock dredge samples supply key information to establish the tectonic and geological framework of the northern two‐thirds of the 95% submerged Zealandia continent. The R/V Investigator voyage IN2016T01 to the Fairway Ridge, Coral Sea, obtained poorly sorted poly‐lithologic pebbly to cobbly sandstones, well sorted fine grained sandstones, mudstones, bioclastic limestones, and basaltic lavas. Post‐cruise analytical work comprised petrography, whole rock geochemical and Sr and Nd isotopic analyses, and U‐Pb zircon, Rb‐Sr, and Ar‐Ar geochronology. A Fairway Ridge cobbly sandstone has a ∼95 Ma (early Late Cretaceous) depositional age; two biotite granite cobbles are 111 ± 1 and 128 ± 1 Ma in age, and some volcanic pebbles are also likely Early Cretaceous. Fairway Ridge basalts have intraplate alkaline chemistry and are of Late Eocene age (∼40–36 Ma). By analogy with South Zealandia, we interpret strong positive continental magnetic anomalies of North Zealandia to mainly result from Late Cretaceous to Cenozoic intraplate basalts, many of them controlled by rifting. A new basement geological map of North Zealandia shows the position of the Mesozoic Gondwana magmatic arc axis (Median Batholith) and other major geological units. This study completes onland and offshore reconnaissance geological mapping of the entire 5 Mkm ² Zealandia continent.

Federated Distillation (FD) extends classic Federated Learning (FL) to a more general training framework that enables model-heterogeneous collaborative learning by Knowledge Distillation (KD) across multiple clients and the server. However, existing KD-based algorithms usually require a set of shared input samples for each client to produce soft-prediction for distillation. Worse still, such a manual selection is accompanied by careful deliberations or prior information on clients' private data distribution, which is not in line with the privacy-preserving characteristic of classic FL. In this paper, we propose a novel training framework to achieve data-independent knowledge transfer by properly designing a distributed generative adversarial network (GAN) between the server and clients that can synthesize shared feature representations to facilitate the FD training. Specifically, we deploy a generator on the server and reuse each local model as a federated discriminator to form a lightweight efficient distributed GAN that can automatically synthesize simulated global feature representations for distillation. Moreover, since the synthesized feature representations are usually more faithful and homologous with global data distribution, faster and better training convergence can be obtained. Extensive experiments on different tasks and heterogeneous models demonstrate the effectiveness of the proposed framework on model accuracy and communication overhead.

In this paper, we propose Hierarchical Federated Learning with Momentum Acceleration (HierMo), a three-tier worker-edge-cloud federated learning algorithm that applies momentum for training acceleration. Momentum is calculated and aggregated in the three tiers. We provide convergence analysis for HierMo, showing a convergence rate of $\mathcal {O}(\frac{1}{T})$ . In the analysis, we develop a new approach to characterize model aggregation, momentum aggregation, and their interactions. Based on this result, we prove that HierMo achieves a tighter convergence upper bound compared with HierFAVG without momentum. We also propose HierOPT, which optimizes the aggregation periods (worker-edge and edge-cloud aggregation periods) to minimize the loss given a limited training time. By conducting the experiment, we verify that HierMo outperforms existing mainstream benchmarks under a wide range of settings. In addition, HierOPT can achieve a near-optimal performance when we test HierMo under different aggregation periods.

Diffusion MRI tractography parcellation classifies streamlines into anatomical fiber tracts to enable quantification and visualization for clinical and scientific applications. Current tractography parcellation methods rely heavily on registration, but registration inaccuracies can affect parcellation and the computational cost of registration is high for large-scale datasets. Recently, deep-learning-based methods have been proposed for tractography parcellation using various types of representations for streamlines. However, these methods only focus on the information from a single streamline, ignoring geometric relationships between the streamlines in the brain. We propose TractCloud, a registration-free framework that performs whole-brain tractography parcellation directly in individual subject space. We propose a novel, learnable, local-global streamline representation that leverages information from neighboring and whole-brain streamlines to describe the local anatomy and global pose of the brain. We train our framework on a large-scale labeled tractography dataset, which we augment by applying synthetic transforms including rotation, scaling, and translations. We test our framework on five independently acquired datasets across populations and health conditions. TractCloud significantly outperforms several state-of-the-art methods on all testing datasets. TractCloud achieves efficient and consistent whole-brain white matter parcellation across the lifespan (from neonates to elderly subjects, including brain tumor patients) without the need for registration. The robustness and high inference speed of TractCloud make it suitable for large-scale tractography data analysis. Our project page is available at https://tractcloud.github.io/.

The measurement of fetal thalamus diameter (FTD) and fetal head circumference (FHC) are crucial in identifying abnormal fetal thalamus development as it may lead to certain neuropsychiatric disorders in later life. However, manual measurements from 2D-US images are laborious, prone to high inter-observer variability, and complicated by the high signal-to-noise ratio nature of the images. Deep learning-based landmark detection approaches have shown promise in measuring biometrics from US images, but the current state-of-the-art (SOTA) algorithm, BiometryNet, is inadequate for FTD and FHC measurement due to its inability to account for the fuzzy edges of these structures and the complex shape of the FTD structure. To address these inadequacies, we propose a novel Swoosh Activation Function (SAF) designed to enhance the regularization of heatmaps produced by landmark detection algorithms. Our SAF serves as a regularization term to enforce an optimum mean squared error (MSE) level between predicted heatmaps, reducing the dispersiveness of hotspots in predicted heatmaps. Our experimental results demonstrate that SAF significantly improves the measurement performances of FTD and FHC with higher intraclass correlation coefficient scores in FTD and lower mean difference scores in FHC measurement than those of the current SOTA algorithm BiometryNet. Moreover, our proposed SAF is highly generalizable and architecture-agnostic. The SAF’s coefficients can be configured for different tasks, making it highly customizable. Our study demonstrates that the SAF activation function is a novel method that can improve measurement accuracy in fetal biometry landmark detection. This improvement has the potential to contribute to better fetal monitoring and improved neonatal outcomes.

While medical data is integral to building robust predictive machine learning models for medical research, obtaining access to medical data is increasingly difficult. The challenges primarily arise from obtaining consent, concerns around the privacy and security of medical data and the technical challenge of migrating what can be huge datasets to a centralised location. As a result, this chapter analyses the question, “How can we make medical data more accessible for medical research whilst addressing the ethical and technical issues around data privacy and data-sharing?” Moreover, this work expands on federated learning that represents a paradigm shift in machine learning from both a technical and sociological perspective. From a technical perspective, federated learning enables machine learning models to be trained in a decentralised manner. It thus allows researchers to utilise data stored in separate locations. From a sociological perspective, federated learning represents a shift in the power dynamic between those providing and those using medical data for research. Under a federated learning framework, raw data never leaves the client’s device. Instead, the centralised server only receives encrypted parameter updates after a shared model is sent and trained locally on each client device. This ensures that the entities that provide medical data have more control over where their data is stored and what information is shared with other parties. Even though federated algorithms have a slightly lower accuracy when compared to non-federated algorithms, it comes with data privacy benefits that non-federated algorithms cannot provide.

Image registration is a fundamental requirement for medical image analysis. Deep registration methods based on deep learning have been widely recognized for their capabilities to perform fast end-to-end registration. Many deep registration methods achieved state-of-the-art performance by performing coarse-to-fine registration, where multiple registration steps were iterated with cascaded networks. Recently, Non-Iterative Coarse-to-finE (NICE) registration methods have been proposed to perform coarse-to-fine registration in a single network and showed advantages in both registration accuracy and runtime. However, existing NICE registration methods mainly focus on deformable registration, while affine registration, a common prerequisite, is still reliant on time-consuming traditional optimization-based methods or extra affine registration networks. In addition, existing NICE registration methods are limited by the intrinsic locality of convolution operations. Transformers may address this limitation for their capabilities to capture long-range dependency, but the benefits of using transformers for NICE registration have not been explored. In this study, we propose a Non-Iterative Coarse-to-finE Transformer network (NICE-Trans) for image registration. Our NICE-Trans is the first deep registration method that (i) performs joint affine and deformable coarse-to-fine registration within a single network, and (ii) embeds transformers into a NICE registration framework to model long-range relevance between images. Extensive experiments with seven public datasets show that our NICE-Trans outperforms state-of-the-art registration methods on both registration accuracy and runtime.

To obtain high-quality positron emission tomography (PET) scans while reducing radiation exposure to the human body, various approaches have been proposed to reconstruct standard-dose PET (SPET) images from low-dose PET (LPET) images. One widely adopted technique is the generative adversarial networks (GANs), yet recently, diffusion probabilistic models (DPMs) have emerged as a compelling alternative due to their improved sample quality and higher log-likelihood scores compared to GANs. Despite this, DPMs suffer from two major drawbacks in real clinical settings, i.e., the computationally expensive sampling process and the insufficient preservation of correspondence between the conditioning LPET image and the reconstructed PET (RPET) image. To address the above limitations, this paper presents a coarse-to-fine PET reconstruction framework that consists of a coarse prediction module (CPM) and an iterative refinement module (IRM). The CPM generates a coarse PET image via a deterministic process, and the IRM samples the residual iteratively. By delegating most of the computational overhead to the CPM, the overall sampling speed of our method can be significantly improved. Furthermore, two additional strategies, i.e., an auxiliary guidance strategy and a contrastive diffusion strategy, are proposed and integrated into the reconstruction process, which can enhance the correspondence between the LPET image and the RPET image, further improving clinical reliability. Extensive experiments on two human brain PET datasets demonstrate that our method outperforms the state-of-the-art PET reconstruction methods. The source code is available at https://github.com/Show-han/PET-Reconstruction.

Wireless capsule endoscopy (WCE) is a painless and non-invasive diagnostic tool for gastrointestinal (GI) diseases. However, due to GI anatomical constraints and hardware manufacturing limitations, WCE vision signals may suffer from insufficient illumination, leading to a complicated screening and examination procedure. Deep learning-based low-light image enhancement (LLIE) in the medical field gradually attracts researchers. Given the exuberant development of the denoising diffusion probabilistic model (DDPM) in computer vision, we introduce a WCE LLIE framework based on the multi-scale convolutional neural network (CNN) and reverse diffusion process. The multi-scale design allows models to preserve high-resolution representation and context information from low-resolution, while the curved wavelet attention (CWA) block is proposed for high-frequency and local feature learning. Moreover, we combine the reverse diffusion procedure to optimize the shallow output further and generate images highly approximate to real ones. The proposed method is compared with eleven state-of-the-art (SOTA) LLIE methods and significantly outperforms quantitatively and qualitatively. The superior performance on GI disease segmentation further demonstrates the clinical potential of our proposed model. Our code is publicly accessible at github.com/longbai1006/LLCaps.

Accurate segmentation of pulmonary airways and vessels is crucial for the diagnosis and treatment of pulmonary diseases. However, current deep learning approaches suffer from disconnectivity issues that hinder their clinical usefulness. To address this challenge, we propose a post-processing approach that leverages a data-driven method to repair the topology of disconnected pulmonary tubular structures. Our approach formulates the problem as a keypoint detection task, where a neural network is trained to predict keypoints that can bridge disconnected components. We use a training data synthesis pipeline that generates disconnected data from complete pulmonary structures. Moreover, the new Pulmonary Tree Repairing (PTR) dataset is publicly available, which comprises 800 complete 3D models of pulmonary airways, arteries, and veins, as well as the synthetic disconnected data. Our code and data are available at https://github.com/M3DV/pulmonary-tree-repairing.

Survival prediction is crucial for cancer patients as it provides early prognostic information for treatment planning. Recently, deep survival models based on deep learning and medical images have shown promising performance for survival prediction. However, existing deep survival models are not well developed in utilizing multi-modality images (e.g., PET-CT) and in extracting region-specific information (e.g., the prognostic information in Primary Tumor (PT) and Metastatic Lymph Node (MLN) regions). In view of this, we propose a merging-diverging learning framework for survival prediction from multi-modality images. This framework has a merging encoder to fuse multi-modality information and a diverging decoder to extract region-specific information. In the merging encoder, we propose a Hybrid Parallel Cross-Attention (HPCA) block to effectively fuse multi-modality features via parallel convolutional layers and cross-attention transformers. In the diverging decoder, we propose a Region-specific Attention Gate (RAG) block to screen out the features related to lesion regions. Our framework is demonstrated on survival prediction from PET-CT images in Head and Neck (H&N) cancer, by designing an X-shape merging-diverging hybrid transformer network (named XSurv). Our XSurv combines the complementary information in PET and CT images and extracts the region-specific prognostic information in PT and MLN regions. Extensive experiments on the public dataset of HEad and neCK TumOR segmentation and outcome prediction challenge (HECKTOR 2022) demonstrate that our XSurv outperforms state-of-the-art survival prediction methods.

Type 2 diabetes (T2D) is a persisting issue affecting millions worldwide. Therefore, developing effective prevention and management strategies for T2D is crucial to improve the health and quality of life of individuals affected by this condition. Due to its significant impact on the public health sector, the researchers focused on developing predictive models to predict T2D and identify potential risk factors associated with the development of the condition. Predictive models for T2D have been proposed using various approaches, including machine learning and deep learning algorithms. Using traditional statistical methods, these models can analyze large datasets and identify patterns that may not be apparent. Some models consider factors such as age, sex, family history, lifestyle habits, and medical history to predict the risk of developing T2D. However, the relationship between meal frequency and blood sugar level remains a controversial research topic. The appropriate meal frequency depends on an individual’s health information and diet habits. Therefore, investigating the hypothesis that meal frequency significantly impacts diabetes prediction, but the relationship is not simply linear, could be helpful. In this study, we aim to build a diabetes predictive model using machine learning methods. We investigate the relationship between meal frequency and the incidence rate of type 2 diabetes using data from the Korean National Health and Nutrition Examination Survey (KNHANES) 2013–2014. Furthermore, we employ seven machine learning and deep learning algorithms to verify the hypothesis. Our experiments reveal the best-performing model, demonstrating a prediction accuracy of approximately 0.818. Interestingly, our results show that meal frequency has low importance. We compare our experimental results with state-of-the-art models and discuss the different conclusions these studies report.