Search

Objective: To explore the application value of applying deep learning (DL) algorithm in the grading assessment of corneal fluorescein staining., Methods: A cross-sectional study was carried out, covering 600 corneal fluorescein staining photos acquired in the Contact Lens Clinic, West China Hospital, Sichuan University between 2020 and 2022. Out of the 600 photos, 500 were used to construct the algorithm and the remaining 100 were used for the validation of the algorithm and a comparative analysis of the difference in grading accuracy (ACC) and the length of diagnostic time between artificial intelligence (AI) and optometry students. One month after finishing the first grading analysis, assessment by AI and optometry students was conducted for a second time and results from the two rounds of assessment were compared to examine the intrarater agreement ( kappa value) of the two analyses. The grading analysis results of 3 experienced optometrists were used as the gold standard in the study., Results: Findings of the cross validation with the complete dataset, the training dataset, and the test dataset showed that ResNet34 had the highest predictive accuracy among four DL models. ResNet34 DL model achieved an accuracy of 93.0%, sensitivity of 89.5%, and specificity of 89.6% in the grading of corneal staining. In the comparison of the grading accuracy of AI and two optometry students, AI showed better accuracy, with the respective grading accuracy being 87.0%, 78.0%, and 52.0% for AI, student 1, and student 2 ( P ACC =0.001). In addition, the average diagnostic time of AI was shorter than that of optometry students ( t AI =1.00 s, t S1 =11.86 s, t S2 =13.25 s, P t =0.001). In the comparative analysis of the intrarater agreement between the two assessments, AI ( kappa AI =0.658, P AI =0.001) achieved better consistency than the two optometry students did （ kappa S1 =0.575, P S1 =0.001; kappa S2 =0.609, P S2 =0.001)., Conclusion: Applying deep learning algorithms in the grading assessment of corneal fluorescein staining has considerable feasibility and clinical value. In the performance comparison between AI and optometry students, AI achieved higher accuracy and better consistency, which indicates that AI has potential application value for assisting optometrists to make clinical decisions with speed and accuracy., (Copyright© by Editorial Board of Journal of Sichuan University (Medical Sciences).)

Single-molecule localization microscopy (SMLM) can be used to resolve subcellular structures and achieve a tenfold improvement in spatial resolution compared to that obtained by conventional fluorescence microscopy. However, the separation of single-molecule fluorescence events that requires thousands of frames dramatically increases the image acquisition time and phototoxicity, impeding the observation of instantaneous intracellular dynamics. Here we develop a deep-learning based single-frame super-resolution microscopy (SFSRM) method which utilizes a subpixel edge map and a multicomponent optimization strategy to guide the neural network to reconstruct a super-resolution image from a single frame of a diffraction-limited image. Under a tolerable signal density and an affordable signal-to-noise ratio, SFSRM enables high-fidelity live-cell imaging with spatiotemporal resolutions of 30 nm and 10 ms, allowing for prolonged monitoring of subcellular dynamics such as interplays between mitochondria and endoplasmic reticulum, the vesicle transport along microtubules, and the endosome fusion and fission. Moreover, its adaptability to different microscopes and spectra makes it a useful tool for various imaging systems., (© 2023. The Author(s).)

This paper delves into leveraging deep learning techniques, such as graph neural networks (GNNs), Transformer, and techniques in Large Language Models (LLMs), to enhance course recommendation systems in e-learning platforms. Recommendation methods have some short-comes in the case of online course with less information and choic less logic. Our research proposes novel algorithms that use graph collaborative filtering and sequential recommendation to improve recommendation accuracy and personalization. By analyzing user behavior patterns and course attributes, our approach aims to provide smarter and more efficient course recommendation services, ultimately enhancing learning outcomes and experiences in e-learning environments. This research not only contributes to the advancement of e-learning technology but also provides valuable insights for the broader application of deep learning in smart education. [ABSTRACT FROM AUTHOR]

Consistent annotation transfer from reference dataset to query dataset is fundamental to the development and reproducibility of single-cell research. Compared with traditional annotation methods, deep learning based methods are faster and more automated. A series of useful single cell analysis tools based on autoencoder architecture have been developed but these struggle to strike a balance between depth and interpretability. Here, we present TOSICA, a multi-head self-attention deep learning model based on Transformer that enables interpretable cell type annotation using biologically understandable entities, such as pathways or regulons. We show that TOSICA achieves fast and accurate one-stop annotation and batch-insensitive integration while providing biologically interpretable insights for understanding cellular behavior during development and disease progressions. We demonstrate TOSICA's advantages by applying it to scRNA-seq data of tumor-infiltrating immune cells, and CD14+ monocytes in COVID-19 to reveal rare cell types, heterogeneity and dynamic trajectories associated with disease progression and severity. Developing computational tools for interpretable cell type annotation in scRNA-seq data remains challenging. Here the authors propose a Transformer-based model for interpretable annotation transfer using biologically understandable entities, and demonstrate its performance on large or atlas datasets. [ABSTRACT FROM AUTHOR]

A localization system is one of the basic requirements for coal mines. Ultra-wideband (UWB), as a technology with broad application prospects, is considered to have great potential in the absence of satellite signals, especially in the underground mine environment, as it can provide rescue assistance. However, state-of-the-art UWB position systems in coal mines cannot efficiently differ the line-of-sight from all communication links, which results in deterioration of the localization accuracy. In this paper, we propose a LOS/NLOS classification method based on a deep learning algorithm. Specifically, we utilize the Generative Adversarial Networks (GAN) to generate diagnostic data for frame transmission under non-line-of-sight (NLOS) condition. Then, a Convolutional Neural Network (CNN) is adopted to identify the NLOS communication. Finally, the trilateral centroid positioning algorithm (TCPA) based on ranging data is used to verify the effect of our method for a localization system in coal mines. Field experiments show that our method can accurately differ the LOS/NLOS with the accuracy of 91.19%. The TCPA algorithm with our method can obtain 3.11% improvement compared with the scenario without using our method. [ABSTRACT FROM AUTHOR]

In face recognition systems, highly robust facial feature representation and good classification algorithm performance can affect the effect of face recognition under unrestricted conditions. To explore the anti-interference performance of convolutional neural network (CNN) reconstructed by deep learning (DL) framework in face image feature extraction (FE) and recognition, in the paper, first, the inception structure in the GoogleNet network and the residual error in the ResNet network structure are combined to construct a new deep reconstruction network algorithm, with the random gradient descent (SGD) and triplet loss functions as the model optimizer and classifier, respectively, and it is applied to the face recognition in Labeled Faces in the Wild (LFW) face database. Then, the portrait pyramid segmentation and local feature point segmentation are applied to extract the features of face images, and the matching of face feature points is achieved using Euclidean distance and joint Bayesian method. Finally, Matlab software is used to simulate the algorithm proposed in this paper and compare it with other algorithms. The results show that the proposed algorithm has the best face recognition effect when the learning rate is 0.0004, the attenuation coefficient is 0.0001, the training method is SGD, and dropout is 0.1 (accuracy: 99.03%, loss: 0.0047, training time: 352 s, and overfitting rate: 1.006), and the algorithm proposed in this paper has the largest mean average precision compared to other CNN algorithms. The correct rate of face feature matching of the algorithm proposed in this paper is 84.72%, which is higher than LetNet-5, VGG-16, and VGG-19 algorithms, the correct rates of which are 6.94%, 2.5%, and 1.11%, respectively, but lower than GoogleNet, AlexNet, and ResNet algorithms. At the same time, the algorithm proposed in this paper has a faster matching time (206.44 s) and a higher correct matching rate (88.75%) than the joint Bayesian method, indicating that the deep reconstruction network algorithm proposed in this paper can be used in face image recognition, FE, and matching, and it has strong anti-interference. [ABSTRACT FROM AUTHOR]

Electrocardiogram (ECG) is a non-invasive tool to measure the heart's electrical activity. ECG signal based automatic heartbeat classification is a critical task for arrhythmia detection and continues to be challenging. While diverse automated classification methods have been developed, they still cannot provide acceptable performance in classifying different heartbeats because of their poor ability to extract abstract patterns comprehensively. Besides, the performance of previous work drops sharply when dealing with imbalanced datasets and lacks interpretability. This paper proposes a novel, explainable attention-based temporal convolutional network(TCN) heartbeat classification method. The first contribution of our approach is that we fuse the TCN architecture and self-attention mechanism to encode the ECG heartbeat sequences. Specifically, TCN and the self-attention block are designed to capture global variation tends and local features, respectively, to best serve the classification. Meanwhile, multi-class focal loss helps model training overcome the class imbalance problem. In the end, the dynamic perturbation based high-fidelity explanation module was introduced to understand the AI-based model and enhance the model's transparency to clinicians. Experiments on the MIT-BIH-AD dataset demonstrate that our model with a simpler architecture can achieve 99.84% accuracy, 99.90% specificity and 99.60% precision for the intra-patient scheme and 87.81% accuracy, 91.85% sensitivity and 89.81% precision for the inter-patient scheme, which outperforms most of the state-of-the-art(SOTA) works, especially for minority classes. • This method can capture heartbeats' global and local features with lightweight network architecture. • A novel loss function is developed to overcome the class imbalance problem. • We introduce an explaining module to study the influence of features at each time on ECG classification results. • Our model outperforms the state-of-the-art methods, and its diagnostic basis conforms to the clinical experience. [ABSTRACT FROM AUTHOR]

In order to overcome the low efficiency of nonlinear dynamic solution, a hybrid deep neural network (HDNN) is constructed based on the integrated deep learning networks for both anticipating nonlinear dynamics as well as evaluating the damping performance of the bladed-disk system in the study. Different physical models are established respectively to characterize the vibratory characteristics and nonlinear solution method is adopted to obtain the true dynamics for surrogate model construction. The results show that HDNN can realize the dynamic prediction with accuracy compared with true numerical solutions under different contact design parameters. The further findings in the case of lumped parameter modeling indicate that high prediction accuracy and good training performance can be obtained by the proposed HDNN when the training size is 0.6 in contrast to the traditional machine learning techniques. The average dynamic prediction error of amplitude and velocity are less than 0.5% and 1.5% respectively, and the average level of damping performance error is almost zero. The training loss can be reduced to 10−4 orders of magnitude to realize the good convergence, and only the cost within 5 m s is required to accomplish the prediction. Furthermore, HDNN-based multi-objective design optimization is conducted by three different algorithms to obtain the Pareto solution and identify the optimal design parameters. The results exhibit that the NSGA-II outperforms the other two algorithms in capturing the Pareto front solutions. And the dynamic predictions are discussed with different optimal candidates obtained by NSGA-II. The proposed HDNN can meet the demand of accurate dynamic and performance prediction as well as efficient design optimization, providing the analytical support for preliminary turbine design in engineering scenarios. • A HDNN framework is proposed for predicting nonlinear dynamics and damping performance of the bladed-disk with friction contact. • High prediction accuracy and little time cost can be obtained by HDNN compared with traditional methods. • The HDNN-based multi-objective design optimization is efficiently carried out to obtain the Pareto solutions. • The dynamic predictions are discussed with different candidates obtained by NSGA-II to identify the optimal design parameters. [ABSTRACT FROM AUTHOR]

Edge devices offer advantages such as low computation latency and high data security for executing convolutional neural networks (CNNs). However, deploying CNNs on resource-constrained devices is challenging due to the high computational intensity of CNNs and limited hardware on-chip resources. This hinders the application of deep learning techniques on edge devices. To address this issue, this paper proposes a reconfigurable CNN edge computing system based on Field-Programmable Gate Array (FPGA) for target detection tasks. The system utilizes the pipeline structure of FPGAs to speed up network computation and employs off-chip memory to store network models. Thus, the need for tiling techniques with high intermediate cache requirements was circumvented in this system. Additionally, we developed a parallel data scheduling model to reduce storage access cost delay on network computation efficiency. Our model achieves comparable efficiency with on-chip storage-based works when using off-chip storage. Online reconfigurable design enables the system to configure network structure and parameters at runtime to achieve different target recognition. This provides greater flexibility for use cases with frequently changing requirements. The proposed system was implemented on a Spartan-6 XC6SLX150 FPGA platform and was applied to pedestrian and vehicle classification tasks. To evaluate performance, speed, power consumption, and average intersection over union (IoU) were separately measured. Our system achieved a detection speed of 16 frames per second (FPS) on the Spartan-6, with a power consumption rate of 0.79 W and an average IoU of 41.2%. Remarkably, the system demonstrated a 178% speed increase and a 60% power consumption reduction compared to the FPGA-based FitNN implementation, while classification accuracy was reduced by only 2.52%. [ABSTRACT FROM AUTHOR]