Search

A recent study proposes an unsupervised neural network-based image stitching method for bladder endoscopy. Traditional image stitching methods rely on feature matching, but this study utilizes deep-learning techniques to overcome the limitations of supervised learning approaches. The method consists of two modules: an unsupervised alignment network and an unsupervised fusion network. Experimental results show a high success rate and accuracy in image stitching, eliminating artifacts and enhancing resolution. This advancement in cystoscopy image stitching could pave the way for real-time panoramic stitching of bladder endoscopic video images and the development of computer-vision-assisted diagnostic systems for bladder cavities. [Extracted from the article]

A patent has been issued to Intel Corporation for a method of compressing deep neural networks for efficient implementation on mobile devices. The method involves disconnecting connections between layers of the network based on the weight of the connection, and then updating the weights through further training. This process generates a sparsely connected deep neural network with fewer connections than the original network. The method also includes generating a connection matrix and applying a discriminative function to determine whether each connection is connected or not. The patent provides a detailed description of the method and its potential applications. [Extracted from the article]

Gauss Surgical Inc. has been issued a patent for methods and systems for processing an image of a machine-readable code that is partially obscured by a substance, such as blood. The patent describes techniques for adjusting the color space of the image, binarizing the machine-readable code region, and decoding the coded information. The methods and systems may be used in various applications, including processing machine-readable codes on surgical textiles. The patent also mentions the use of trained neural networks and image processing algorithms to improve the accuracy of the image processing. [Extracted from the article]

Cosmo Artificial Intelligence-ai Limited has been issued a patent for systems and methods for processing real-time video from a medical image device and detecting objects in the video. The patent describes a computer-implemented system that includes an object detector network and an adversarial network for distinguishing true positives from false positives. The system combines neural networks to improve object detection and location information. It also allows for real-time video display with object detections and provides options for adjusting sensitivity settings and weights of the trained networks. The patent was filed in 2019 and published in 2024. [Extracted from the article]

A recent study conducted by researchers from the University of Gour Banga in Malda, India, explores the use of advanced computational techniques in medical diagnostics and treatment planning, specifically in angioplasty procedures. The study focuses on the use of balloon catheters in a bloodstream containing tetra-hybrid nanoparticles within a stenotic arterial cavity influenced by a magnetic field. The researchers developed a model to examine factors such as blood flow, temperature, pressure gradient, and heat transfer coefficient. The study found that increasing the height of the balloon catheter increases blood velocity, while a thicker interfacial nanolayer decreases blood temperature. The researchers also developed an artificial neural network model to accurately predict wall shear stress and heat transfer coefficient. The study concludes that this research could lead to more targeted and efficient use of balloon catheters in treating arterial blockages, benefiting cardiovascular medicine and patient care. [Extracted from the article]

A recent report from the University of Sao Paulo discusses the use of a Python-based algorithm and the MediaPipe library in Maxillofacial Surgery. The advent of FaceMesh, a deep neural network integrated with Virtual Surgical Planning, has revolutionized the field by providing precise 4D facial mapping capabilities. This technology allows surgeons to tailor interventions to individual patients, streamline surgical procedures, and improve aesthetic and functional results. Ongoing refinement and validation are necessary to ensure the reliability and effectiveness of FaceMesh in clinical settings. [Extracted from the article]

A study conducted at the Federal University Santa Catarina in Florianopolis, Brazil, has explored the use of spiking neural networks in controlling upper limb prostheses. The researchers found that a simple two-layered architecture with a few tens of neurons was able to accurately detect touch and slippage with nearly 100% accuracy. They also discovered that implementing the network in ultralow-power neuromorphic hardware could significantly reduce power consumption. This research has the potential to enhance the dexterity and embodiment of prostheses for patients, addressing issues such as prosthesis abandonment. [Extracted from the article]

A patent has been issued to Leica Biosystems Imaging Inc. for a method of processing pathology images using a neural network. The method involves using a convolutional neural network to identify areas of clinical interest and areas of toxicological pathology in histological images. The neural network is trained using a dataset of histological images and pathologist interaction data. The method also includes generating a toxicity map and analyzing it along with a segmentation map to identify areas of interest that exhibit toxicity. The patent also describes an apparatus and method for visualizing and selecting toxic areas in the histological images. [Extracted from the article]

The patent application titled "Medical Image Augmentation" proposes a computerized method and apparatus for improving ultrasound imaging of bony anatomies. The invention involves using machine learning to generate synthetic images that depict anatomical structures not visible or distorted in the source images. These synthetic images are then combined with the source images to create augmented digital images. The invention aims to enhance the interpretation of medical images, particularly for diagnostic and interventional procedures. [Extracted from the article]

A study conducted at Wuhan University in China has developed a new noninvasive technique for automatically diagnosing adenomyosis using deep neural networks. The technique, called A2DNet, achieved an accuracy of 92.33% in diagnosing adenomyosis based on ultrasound images. The researchers concluded that this technique, which utilizes artificial intelligence, is unique and advantageous due to its nondestructive and non-invasive nature. The study was published in the European Journal of Obstetrics & Gynecology and Reproductive Biology. [Extracted from the article]

A report from the Harbin Institute of Technology in Heilongjiang, China discusses the importance of high-resolution Magnetic Resonance Imaging (MRI) in diagnosing and treating brain tumors. The researchers propose a novel network architecture that utilizes convolutional neural networks to improve the acquisition of high-resolution MRI images. They conducted tests on four datasets and found that their method outperformed current leading-edge methods in terms of accuracy. The research was supported by the National Key R&D Program of China and has been peer-reviewed. [Extracted from the article]

A recent report from the University of Sevilla in Spain highlights the importance of early and accurate detection of brain tumors for improving patient outcomes. The researchers focused on developing an image classifier using convolutional neural networks (CNN) to detect brain tumors in magnetic resonance imaging (MRI). The study found that CNNs achieved exceptional performance in analyzing medical images, surpassing previous studies in accuracy and reliability. These findings demonstrate the potential of deep learning techniques in clinical applications, enhancing diagnostic capabilities. [Extracted from the article]

Researchers from King Abdulaziz University in Saudi Arabia have developed a new deep learning-based model called ODL-BAAM for Bone Age Assessment (BAA) using X-ray images of hands. The model utilizes preprocessing steps to standardize and normalize the X-ray data, and integrates Faster RCNN with MobileNet for more effective feature extraction. The Tunicate Swarm Algorithm is used to optimize model hyperparameters, and age determination is facilitated through SoftMax layers. Extensive simulation studies have shown that ODL-BAAM outperforms existing BAA models, achieving a remarkable 96.5% accuracy rate. This research represents a significant advancement in computerized BAA and sets a new standard for medical image analysis. [Extracted from the article]

A research study conducted at Fudan University in Shanghai, China, has developed a pseudo-siamese neural network called PSEENet that incorporates electroencephalography (EEG) and electrooculography (EOG) characteristics for sleep staging. Sleep staging is important for evaluating the quality of sleep, but current studies struggle with varying input modalities and heterogeneous signals. PSEENet consists of spatial mapping modules and a weight-shared classifier to handle these challenges. The model was evaluated on public datasets and a clinical dataset, showing superior results with multimodal signals and good performance with single modality. The research has been peer-reviewed and published in the IEEE Journal of Biomedical and Health Informatics. [Extracted from the article]

Researchers at Ewha Womans University in Seoul, South Korea have developed a new method to enhance the visualization of arteries in contrast-enhanced cerebral MR angiography (MRA) using generative neural networks. The study aimed to generate pure angiograms with enhanced arterial contrast by utilizing contrast dynamics. The results showed that the proposed method achieved higher arterial contrast compared to raw MRA data. The researchers also demonstrated the feasibility of applying the trained generative models to single-phase contrast-enhanced MRA images with high spatial resolution. This research has been peer-reviewed and published in Biomedical Signal Processing and Control. [Extracted from the article]

Researchers at Purdue University have developed a new vaccine prioritization strategy to reduce the burden of the pandemic when vaccine supply is limited. The study incorporates mobility patterns and uses deep reinforcement learning to optimize vaccine allocation. The proposed framework reduces infections and deaths by 7%-10% compared to baseline strategies and is robust in diverse mobility patterns. The research provides valuable insights for areas with limited vaccines and low logistic efficacy. [Extracted from the article]

Amazon Technologies Inc. has been issued a patent for updating machine learning models across devices. The patent describes a system and method for adapting a global model to accommodate personalized local models, balancing globalization and localization based on variance/uncertainty data. The system allows for the customization of machine learning models for local operation without straying too far from the global model. The patent also emphasizes the importance of user permissions and compliance with laws and regulations. [Extracted from the article]

Researchers at Sichuan University in China have developed an injectable double-crosslinked collagen/hyaluronic acid hydrogel loaded with nerve growth factor (NGF) that can repair irregular bone defects. The hydrogel demonstrated excellent properties such as injectability, self-healing, and rapid crosslinking, along with suitable mechanics, rheology, swelling, degradation, and biocompatibility. In vitro and in vivo experiments showed that the hydrogel promoted neuron growth and osteogenic differentiation, leading to better bone repair compared to other materials. This research provides a promising material for the functional and structural repair of complex bone defects. [Extracted from the article]

A study conducted by researchers at Islamic Azad University in Esfahan, Iran, focused on the synthesis and characterization of a biodegradable magnetic nanocomposite scaffold for potential bone tissue engineering applications. The researchers incorporated polyvinyl alcohol with hydroxyapatite nanoparticles and iron oxide nanoparticles using a freeze-drying technique. They evaluated the physicochemical and morphological properties of the scaffold and found impressive mechanical characteristics, such as a compressive strength of 4.61 MPa and an elastic modulus of 2.13 GPa. The researchers also developed an artificial neural network (ANN) to predict and optimize the properties of the scaffold. The study concluded that the synthesized scaffold closely resembles the physical and mechanical characteristics of natural bone tissue. [Extracted from the article]

Researchers from Tianjin University in China have developed a neural networks-based smart e-health application for the prediction of tuberculosis (TB) using serverless computing. The study evaluated the performance of various convolution neural network (CNN) architectures and found that integrating advanced machine learning models within IoT and e-health frameworks could enable more comprehensive and data-driven approaches to healthcare. The best-performing model, VGG-19, was selected for different deployment strategies using server and serverless-based environments. The research suggests that integrating these models into IoT and e-health systems could enhance healthcare outcomes through more informed and timely treatments. [Extracted from the article]

A recent report from researchers at Sultan Qaboos University in Muscat, Oman, highlights the importance of effective patient classification and resource allocation in healthcare systems, particularly in the context of pandemics like COVID-19. The study proposes a patient clustering method using Artificial Neural Networks (ANN) and Data Envelopment Analysis-Discriminant Analysis (DEA-DA) to assign patients to appropriate treatment facilities based on specific testing criteria. The approach aims to improve healthcare system efficiency by ensuring that patients with severe illnesses receive care at well-equipped centers, while those with milder symptoms are directed to other suitable facilities. The research concludes that this method has the potential to increase recovery rates and decrease mortality rates. [Extracted from the article]

A recent study conducted at the University of Connecticut explores the relationship between protein dynamics and protein crystallization propensity. The researchers propose a graph neural network model that incorporates protein dynamics to predict the likelihood of protein crystallization. By comparing their results to molecular dynamics simulations, the researchers found that fluctuating residues in proteins correlate with high attention scores in the neural network. This study demonstrates the integration of physics-based and machine learning models in structural biology, offering insights for the rational design of protein sequences. [Extracted from the article]

Researchers from the Chinese University of Hong Kong Shenzhen have developed a new method for classifying nuclei in histopathology images. The researchers propose a novel polygon-structure feature learning mechanism that transforms a nucleus contour into a sequence of points, and employ a recurrent neural network to obtain learnable shape features. They also convert histopathology images into a graph structure with nuclei as nodes and use a graph neural network to embed the spatial distribution of nuclei into their representations. The proposed framework shows significant improvements compared to previous methods. This research has been peer-reviewed and published in IEEE Transactions on Medical Imaging. [Extracted from the article]

A study conducted by researchers at Nanjing University of Aeronautics and Astronautics in China has proposed a bi-Gamma distribution model to determine the probability density function (PDF) of broadband non-Gaussian random stress rainflow ranges during vibration fatigue. The researchers used a neural network to establish a model that can provide the parameter values of the bi-Gamma model based on input PSD parameters and kurtosis, thereby obtaining the PDF of the stress rainflow ranges. The study verified the predictive capability of the neural network model and compared the effects of non-Gaussian random stress with different kurtosis on structural fatigue life. The results showed that all life predicted results were within the second scatter band. [Extracted from the article]

A recent study conducted by researchers at the University of Southern Florida has found that deep learning techniques can enhance the detection of Alzheimer's disease. The researchers utilized transfer learning, which involves leveraging pre-trained models from similar tasks, and data augmentation techniques to increase the dataset size. They applied these techniques to magnetic resonance imaging (MRI) datasets and achieved high precision, recall, F1 score, and accuracy for Alzheimer's disease detection. The study concludes that this combination of transfer learning and ensemble learning represents a significant advancement in medical diagnostics. [Extracted from the article]

A recent study conducted at the University of Malaya in Kuala Lumpur, Malaysia, focused on the development of a new model for classifying gastric histopathology images using deep learning techniques. The researchers designed a dilated convolutional neural network with a late fusion strategy (DCNNLFS) to improve the accuracy of classification. The DCNNLFS model utilizes dilated convolutions to expand the receptive field and learn contextual information. Experimental results showed high precision, accuracy, and F1-Score for the DCNNLFS model. This research contributes to the field of engineering and has potential implications for improving the diagnosis of gastric diseases. [Extracted from the article]

Researchers from Korea University in Seoul, South Korea have developed a new method for decoding electroencephalography (EEG) signals using a learnable continuous wavelet-based multi-branch attentive convolutional neural network. This method aims to improve the extraction of spatio-spectral-temporal (SST) features from EEG signals, which is important for motor imagery (MI) classification. The researchers found that their proposed method outperformed existing methods and could have implications for personalized EEG decoding. This research has been peer-reviewed and published in Expert Systems with Applications. [Extracted from the article]

A study conducted at Bar Ilan University in Israel has developed a new methodology for classifying cognitive states using convolutional neural networks (CNNs) on electroencephalography (EEG) data. The study focused on analyzing EEG signals recorded from the inferior parietal lobule during various cognitive tasks and achieved an accuracy of 91.42% in distinguishing between high and low gamma activity states. This approach offers significant time savings over manual analysis methods and has implications for brain-computer interfaces, clinical diagnostics, and cognitive monitoring. The research was published in the journal Applied Sciences. [Extracted from the article]

A study conducted by researchers at Oregon Health & Science University (OHSU) has developed and evaluated machine learning models to improve the accuracy and efficiency of trachoma surveys. Trachoma surveys are currently conducted by human graders, which can be resource-intensive and prone to human error. The study used deep convolutional neural networks to detect cases of trachomatous inflammation-follicular (TF) and the number of follicles evident in conjunctival photographs. The models performed well in classifying TF and detecting follicles, suggesting that similar models could enable accurate and large-scale trachoma screening. Further validation in diverse populations is needed before implementation at scale. [Extracted from the article]

A study from Guangzhou University in China explores the use of electroencephalography (EEG) for emotion recognition. The researchers propose a new framework that combines a temporal convolutional network (TCN) and a 3D convolutional neural network (3DCNN) to learn the frequency band relationships and temporal-spatial representation of EEG data. The framework achieves high accuracy in emotion recognition compared to current methods. The study also investigates the impact of different frequency band divisions and visualizes the extracted features for better interpretability. [Extracted from the article]

Researchers at Indiana University Bloomington have been studying functional neural networks in human brain organoids, which are 3-dimensional brain-like tissues derived from human pluripotent stem cells. While the molecular and cellular aspects of these organoids have been extensively studied, their functional properties, such as organoid neural networks (ONNs), have been largely understudied. The researchers discuss recent advances in understanding, characterizing, and applying functional ONNs in human brain organoids, including the use of microelectrode array technology and calcium imaging. They also highlight how ONNs have been used to investigate neurological diseases like Rett syndrome and Alzheimer's disease. The researchers provide their perspectives on the future challenges and opportunities for using ONNs in basic research and translational applications. [Extracted from the article]

A study conducted by researchers at Southern Medical University in Guangdong, China, has developed a new method for magnetic resonance electrical property tomography (MREPT) using physics-informed neural networks. MREPT is a noninvasive technique that can determine the electrical properties of tissues from measured radiofrequency signals in MRI systems. The researchers proposed a model-driven method that uses fully connected networks to learn the mapping from measured radiofrequency signals to electrical properties. The method was tested using a three-dimensional head model, digital phantom simulation, and a practical phantom experiment. The research provides a novel approach to MREPT and has potential applications in medical imaging. [Extracted from the article]

A recent study conducted by researchers from Peking University in Beijing, China, explored the representation alignment between deep convolutional neural networks (DCNNs) and the human brain in the context of action recognition. The study found that while DCNN models achieved human-level performance for object categorization, they did not demonstrate hierarchical representation alignment with the human brain when processing action videos. Instead, the later layers of DCNN models showed greater representation similarities to the human visual cortex. These findings suggest fundamental differences in how DCNNs and the human brain represent dynamic visual information related to actions. [Extracted from the article]

New research from Shanghai University explores a new type of threat called a backdoor attack on graph convolutional networks (GCNs), which are used for graph-related tasks. The study focuses on semantic backdoor attacks, where a naturally occurring semantic feature of samples can serve as a trigger to misclassify testing samples. The researchers propose a semantic backdoor attack against GCNs (SBAG) and evaluate its effectiveness on five graph datasets. The experimental results show that SBAG can achieve high attack success rates on both unmodified and modified testing samples. This research provides insights into the security vulnerabilities of GCNs. [Extracted from the article]

A recent study conducted by researchers at Jilin Agricultural University in China has proposed a new method for recognizing wild mushrooms based on a compact bilinear neural network. The method utilizes a dual-stream structure and multiple feature extractors to enhance the accuracy of image recognition. Experimental results showed that the proposed method achieved an accuracy of 98.03%, outperforming existing methods. The researchers suggest that the development of public-facing identification tools based on this method could help reduce the risk of poisoning incidents. They also recommend further research and development of digital agricultural technologies to promote the application of intelligent recognition technologies in agriculture. [Extracted from the article]

A study conducted at Xidian University in Shaanxi, China, has proposed a new approach to improve human visual target detection based on electroencephalography (EEG) signals. The researchers developed a method called the ensemble multi-scale convolution neural network (E_MS_CNN), which extracts multi-scale spatial-temporal features of EEG signals using stacked multi-scale convolutional neural networks. The E_MS_CNN method outperformed existing methods in cross-subject and cross-session situations, according to extensive experimental results. This research has been peer-reviewed and published in Biomedical Signal Processing and Control. [Extracted from the article]

Researchers at Nanjing University of Information Science and Technology (NUIST) in China have developed a new method for detecting and diagnosing small breast masses using convolutional neural networks. The method, called RCM-YOLO, improves the identification of small masses by increasing the resolution of feature maps and optimizing the amount of parameters. The researchers found that RCM-YOLO had a mean Average Precision (mAP) of 90.34% and reduced the missed detection rate for small masses to 11%. This method has the potential to improve the detection rate of small masses and reduce misdiagnosis in breast cancer screening. [Extracted from the article]

A patent application has been published online for a system that aims to provide a less invasive and more accurate test for determining the chromosomal makeup of mammal embryos, including humans. The system uses polarized light image data and a convolutional neural network (CNN) to classify the ploidy status of the embryos. By analyzing the image data, the CNN generates a classification metric that indicates the likelihood of the ploidy status. This technology has the potential to optimize the selection of embryos for implantation, increasing the chances of a successful live birth. The patent application provides detailed technical information about the system and its components. [Extracted from the article]

Researchers at Huazhong University of Science and Technology in Wuhan, China have developed deep learning-driven multimodal information fusion frameworks for lower limb motion recognition. By leveraging the power of deep learning, the researchers constructed several end-to-end frameworks, achieving high accuracy in recognizing different types of lower limb motions in healthy subjects and patients with stroke. The proposed frameworks show promise in supporting the development of human-robot collaborative lower limb exoskeletons or rehabilitation robots. The research has been peer-reviewed and published in Biomedical Signal Processing and Control. [Extracted from the article]

Inventor William Keustermans has filed a patent application for a method of modeling specific nasal cavity shapes using computational fluid dynamics (CFD). The method involves obtaining measurement data of the specific nasal cavity, such as rhinomanometry or acoustic rhinometry data, and inputting it into a neural network to generate a model. This model can be used to improve nasal drug delivery devices and assess nasal airway obstruction before surgery. The invention aims to provide a cost-effective and non-invasive alternative to tomographic imaging for modeling nasal cavity shapes. The patent application also includes claims for various computer-implemented methods and products related to the modeling of nasal cavity shapes. [Extracted from the article]

A patent has been issued for caching techniques for a deep learning accelerator. The device described in the patent is designed to perform computations of artificial neural networks (ANNs) with reduced energy consumption and computation time. It includes a deep learning accelerator (DLA), random access memory, and a system buffer. The DLA has local memory and can access the local memory faster than the system buffer, and the system buffer can access the random access memory faster than the DLA. The patent provides instructions for loading data from the random access memory into the local memory, including hints on the type of data and processing that will be applied in the DLA. The device is programmable to perform the computations of an ANN, with each neuron in the ANN receiving a set of inputs. The patent includes various claims related to the device and its operation. The patent was filed by inventors from different locations in the United States and is assigned to Micron Technology Inc. [Extracted from the article]

Researchers from Tianjin University in China have developed a new method for diagnosing Major Depressive Disorder (MDD) using dynamic-static fusion graph neural networks. MDD is a complex mental condition with unclear diagnostic mechanisms, but research has linked it to abnormal brain connectivity. The new framework, called DSFGNN, addresses limitations in existing models by incorporating a graph isomorphism encoder, a spatiotemporal attention mechanism, and additional modules to enhance interpretability and expressiveness. The researchers evaluated DSFGNN on a dataset and found that it outperformed the best baseline model, showing potential for biomarker discovery. This research has been peer-reviewed and published in the IEEE Journal of Biomedical and Health Informatics. [Extracted from the article]

A research report from the National Institute of Technology in Patna, India discusses findings on diabetic retinopathy (DR), a leading cause of blindness worldwide. The report highlights the development of a data-driven 2D-EWT based automated DR diagnosis system using a hybrid neural network. The system achieved high accuracy rates on three different datasets, outperforming other methods in terms of accuracy, area under the curve, precision, specificity, and sensitivity. The research provides valuable insights into the early detection and diagnosis of DR, offering potential advancements in the field of ophthalmology. [Extracted from the article]

A recent study conducted at the University of Malaya in Kuala Lumpur, Malaysia, explores the use of an innovative approach to improve the detection of diabetic retinopathy. The study integrates the Inception-V4 deep learning-based neural network with a modified dynamic Snow Leopard Optimization (DSLO) algorithm to enhance the accuracy and efficiency of detection. By analyzing digital images obtained during routine eye exams, automated image processing algorithms can identify early signs of diabetic retinopathy, such as leaking vessels or optic nerve edema. The proposed model demonstrated superior performance and improved detection of early-stage diabetic retinopathy cases compared to other state-of-the-art models. [Extracted from the article]

A study conducted at Qufu Normal University in Shandong, China, explores the application of graph neural networks (GNNs) in handling scale-free graph-structured data. The researchers propose a new framework called the Hyperbolic Hypergraph Convolutional Neural Network (HHGNN), which maps scale-free graphs from Euclidean space to hyperbolic space using a Variational Autoencoder (VAE). The HHGNN model demonstrates strong scalability and outperforms current methods in node classification tasks, particularly on datasets with hierarchical structures. The study also includes an analysis of a diabetes dataset to support early diagnosis. This research has been peer-reviewed and published in the journal Neurocomputing. [Extracted from the article]

Researchers from the Norwegian University of Science and Technology (NTNU) have conducted a study on the impact of topology on neural network function. Using microfluidic technologies, they created modular circuit motifs that promote both segregation and integration of information processing in engineered neural networks. The researchers demonstrated that the resulting laminar cortical networks exhibit segregated and integrated functional dynamics across layers, similar to those observed in the neocortex. This model system has the potential to provide new insights into the development, organization, and neuroplasticity mechanisms of the neocortex in both healthy and pathological conditions. [Extracted from the article]