Atrial fibrillation (AF) is the most prevalent form of arrhythmia, with a rising incidence and prevalence worldwide, posing significant implications for public health. In this paper, we introduce an approach that combines the Recurrence Plot (RP) technique and the ResNet architecture to predict AF. Our method involves three main steps: using wavelet filtering to remove noise interference; generating RPs through phase space reconstruction; and employing a multi-level chained residual network for AF prediction. To validate our approach, we established a comprehensive database consisting of electrocardiogram (ECG) recordings from 1008 AF patients and 48,292 Non-AF patients, with a total of 2067 and 93,129 ECGs, respectively. The experimental results demonstrated high levels of prediction precision (90.5%), recall (89.1%), F1 score (89.8%), accuracy (93.4%), and AUC (96%) on our dataset. Moreover, when tested on a publicly available AF dataset (AFPDB), our method achieved even higher prediction precision (94.8%), recall (99.4%), F1 score (97.0%), accuracy (97.0%), and AUC (99.7%). These findings suggest that our proposed method can effectively extract subtle information from ECG signals, leading to highly accurate AF predictions.

Sleep, or the lack thereof, has far-reaching consequences on many aspects of human physiology, cognitive performance, and emotional wellbeing. To ensure undisturbed sleep monitoring, unobtrusive measurements such as ballistocardiogram (BCG) are essential for sustained, real-world data acquisition. Current analysis of BCG data during sleep remains challenging, mainly due to low signal-to-noise ratio, physical movements, as well as high inter- and intra-individual variability. To overcome these challenges, this work proposes a novel approach to improve J-peak extraction from BCG measurements using a supervised deep learning setup. The proposed method consists of the modeling of the discrete reference heartbeat events with a symmetric and continuous kernel-function, referred to as surrogate signal. Deep learning models approximate this surrogate signal from which the target heartbeats are detected. The proposed method with various surrogate signals is compared and evaluated with state-of-the-art methods from both signal processing and machine learning approaches. The BCG dataset was collected over 17 nights using inertial measurement units (IMUs) embedded in a mattress, together with an ECG for reference heartbeats, for a total of 134 h. Moreover, we apply for the first time an evaluation metric specialized for the comparison of event-based time series to assess the quality of heartbeat detection. The results show that the proposed approach demonstrates superior accuracy in heartbeat estimation compared to existing approaches, with an MAE (mean absolute error) of 1.1 s in 64-s windows and 1.38 s in 8-s windows. Furthermore, it is shown that our novel approach outperforms current methods in detecting the location of heartbeats across various evaluation metrics. To the best of our knowledge, this is the first approach to encode temporal events using kernels and the first systematic comparison of various event encodings for event detection using a regression-based sequence-to-sequence model.

The analysis of medical images (MI) is an important part of advanced medicine as it helps detect and diagnose various diseases early. Classifying brain tumors through magnetic resonance imaging (MRI) poses a challenge demanding accurate models for effective diagnosis and treatment planning. This paper introduces AG-MSTLN-EL, an attention-aided multi-source transfer learning ensemble learning model leveraging multi-source transfer learning (Visual Geometry Group ResNet and GoogLeNet), attention mechanisms, and ensemble learning to achieve robust and accurate brain tumor classification. Multi-source transfer learning allows knowledge extraction from diverse domains, enhancing generalization. The attention mechanism focuses on specific MRI regions, increasing interpretability and classification performance. Ensemble learning combines k-nearest neighbor, Softmax, and support vector machine classifiers, improving both accuracy and reliability. Evaluating the model\'s performance on a dataset with 3064 brain tumor MRI images, AG-MSTLN-EL outperforms state-of-the-art models in terms of all classification measures. The model\'s innovative combination of transfer learning, attention mechanism, and ensemble learning provides a reliable solution for brain tumor classification. Its superior performance and high interpretability make AG-MSTLN-EL a valuable tool for clinicians and researchers in medical image analysis.

Autism Spectrum Disorder (ASD) is a heterogeneous disorder in children, and the current clinical diagnosis is accomplished using behavioral, cognitive, developmental, and language metrics. These clinical metrics can be imperfect measures as they are subject to high test-retest variability, and are influenced by assessment factors such as environment, social structure, or comorbid disorders. Advances in neuroimaging coupled with machine-learning provides an opportunity to develop methods that are more quantifiable, and reliable than existing clinical techniques. In this paper, we design and develop a deep-learning model that operates on functional magnetic resonance imaging (fMRI) data, and can classify between ASD and neurotypical brains. We introduce a novel strategy to transform time-series data extracted from fMRI signals into Gramian Angular Field (GAF) while locking in the temporal and spatial patterns in the data. Our motivation is to design and develop a novel framework that could encode the time-series, acquired from fMRI data, into images that can be used by deep-learning architectures that have been successful in computer vision. In our proposed framework called ASD-GResTM, we used a Convolutional Neural Network (CNN) to extract useful features from GAF images. We then used a Long Short-Term Memory (LSTM) layer to learn the activities between the regions. Finally, the output representations of the last LSTM layer are applied to a single-layer perceptron (SPL) to get the final classification. Our extensive experimentation demonstrates high accuracy across 4 centers, and outperforms state-of-the-art models on two centers with an increase in the accuracy of 17.58% and 6.7%, respectively as compared to the state of the art. Our model achieved the maximum accuracy of 81.78% with high degree of sensitivity and specificity. All training, validation, and testing was accomplished using openly available ABIDE-I benchmarking dataset.

The spindle rotation error of computer numerical control (CNC) equipment directly reflects the machining quality of the workpiece and is a key indicator reflecting the performance and reliability of CNC equipment. Existing rotation error prediction methods do not consider the importance of different sensor data. This study developed an adaptive weighted deep residual network (ResNet) for predicting spindle rotation errors, thereby establishing accurate mapping between easily obtainable vibration information and difficult-to-obtain rotation errors. Firstly, multi-sensor data are collected by a vibration sensor, and Short-time Fourier Transform (STFT) is adopted to extract the feature information in the original data. Then, an adaptive feature recalibration unit with residual connection is constructed based on the attention weighting operation. By stacking multiple residual blocks and attention weighting units, the data of different channels are adaptively weighted to highlight important information and suppress redundancy information. The weight visualization results indicate that the adaptive weighted ResNet (AWResNet) can learn a set of weights for channel recalibration. The comparison results indicate that AWResNet has higher prediction accuracy than other deep learning models and can be used for spindle rotation error prediction.

Electroencephalogram (EEG) plays a pivotal role in the detection and analysis of epileptic seizures, which affects over 70 million people in the world. Nonetheless, the visual interpretation of EEG signals for epilepsy detection is laborious and time-consuming. To tackle this open challenge, we introduce a straightforward yet efficient hybrid deep learning approach, named ResBiLSTM, for detecting epileptic seizures using EEG signals. Firstly, a one-dimensional residual neural network (ResNet) is tailored to adeptly extract the local spatial features of EEG signals. Subsequently, the acquired features are input into a bidirectional long short-term memory (BiLSTM) layer to model temporal dependencies. These output features are further processed through two fully connected layers to achieve the final epileptic seizure detection. The performance of ResBiLSTM is assessed on the epileptic seizure datasets provided by the University of Bonn and Temple University Hospital (TUH). The ResBiLSTM model achieves epileptic seizure detection accuracy rates of 98.88-100% in binary and ternary classifications on the Bonn dataset. Experimental outcomes for seizure recognition across seven epilepsy seizure types on the TUH seizure corpus (TUSZ) dataset indicate that the ResBiLSTM model attains a classification accuracy of 95.03% and a weighted F1 score of 95.03% with 10-fold cross-validation. These findings illustrate that ResBiLSTM outperforms several recent deep learning state-of-the-art approaches.

In the context of deep learning models, attention has recently been paid to studying the surface of the loss function in order to better understand training with methods based on gradient descent. This search for an appropriate description, both analytical and topological, has led to numerous efforts in identifying spurious minima and characterize gradient dynamics. Our work aims to contribute to this field by providing a topological measure for evaluating loss complexity in the case of multilayer neural networks. We compare deep and shallow architectures with common sigmoidal activation functions by deriving upper and lower bounds for the complexity of their respective loss functions and revealing how that complexity is influenced by the number of hidden units, training models, and the activation function used. Additionally, we found that certain variations in the loss function or model architecture, such as adding an ℓ2 regularization term or implementing skip connections in a feedforward network, do not affect loss topology in specific cases.

Traditional Chinese medicine (TCM) has relied on pulse diagnosis as a cornerstone of healthcare assessment for thousands of years. Despite its long history and widespread use, TCM pulse diagnosis has faced challenges in terms of diagnostic accuracy and consistency due to its dependence on subjective interpretation and theoretical analysis. This study introduces an approach to enhance the accuracy of TCM pulse diagnosis for diabetes by leveraging the power of deep learning algorithms, specifically LeNet and ResNet models, for pulse waveform analysis. LeNet and ResNet models were applied to analyze TCM pulse waveforms using a diverse dataset comprising both healthy individuals and patients with diabetes. The integration of these advanced algorithms with modern TCM pulse measurement instruments shows great promise in reducing practitioner-dependent variability and improving the reliability of diagnoses. This research bridges the gap between ancient wisdom and cutting-edge technology in healthcare. LeNet-F, incorporating special feature extraction of a pulse based on TMC, showed improved training and test accuracies (73% and 67%, respectively, compared with LeNet\'s 70% and 65%). Moreover, ResNet models consistently outperformed LeNet, with ResNet18-F achieving the highest accuracy (82%) in training and 74% in testing. The advanced preprocessing techniques and additional features contribute significantly to ResNet18-F\'s superior performance, indicating the importance of feature engineering strategies. Furthermore, the study identifies potential avenues for future research, including optimizing preprocessing techniques to handle pulse waveform variations and noise levels, integrating additional time-frequency domain features, developing domain-specific feature selection algorithms, and expanding the scope to other diseases. These advancements aim to refine traditional Chinese medicine pulse diagnosis, enhancing its accuracy and reliability while integrating it into modern technology for more effective healthcare approaches.

Sign language recognition technology can help people with hearing impairments to communicate with non-hearing-impaired people. At present, with the rapid development of society, deep learning also provides certain technical support for sign language recognition work. In sign language recognition tasks, traditional convolutional neural networks used to extract spatio-temporal features from sign language videos suffer from insufficient feature extraction, resulting in low recognition rates. Nevertheless, a large number of video-based sign language datasets require a significant amount of computing resources for training while ensuring the generalization of the network, which poses a challenge for recognition. In this paper, we present a video-based sign language recognition method based on Residual Network (ResNet) and Long Short-Term Memory (LSTM). As the number of network layers increases, the ResNet network can effectively solve the granularity explosion problem and obtain better time series features. We use the ResNet convolutional network as the backbone model. LSTM utilizes the concept of gates to control unit states and update the output feature values of sequences. ResNet extracts the sign language features. Then, the learned feature space is used as the input of the LSTM network to obtain long sequence features. It can effectively extract the spatio-temporal features in sign language videos and improve the recognition rate of sign language actions. An extensive experimental evaluation demonstrates the effectiveness and superior performance of the proposed method, with an accuracy of 85.26%, F1-score of 84.98%, and precision of 87.77% on Argentine Sign Language (LSA64).

UNASSIGNED: To develop a semi-automatic model integrating radiomics, deep learning, and clinical features for Bone Metastasis (BM) prediction in prostate cancer (PCa) patients using Biparametric MRI (bpMRI) images.
UNASSIGNED: A retrospective study included 414 PCa patients (BM, n=136; NO-BM, n=278) from two institutions (Center 1, n=318; Center 2, n=96) between January 2016 and December 2022. MRI scans were confirmed with BM status via PET-CT or ECT pre-treatment. Tumor areas on bpMRI images were delineated as tumor\'s region of interest (ROI) using auto-delineation tumor models, evaluated with Dice similarity coefficient (DSC). Samples were auto-sketched, refined, and used to train the ResNet BM prediction model. Clinical, radiomics, and deep learning data were synthesized into the ResNet-C model, evaluated using receiver operating characteristic (ROC).
UNASSIGNED: The auto-segmentation model achieved a DSC of 0.607. Clinical BM prediction\'s internal validation had an accuracy (ACC) of 0.650 and area under the curve (AUC) of 0.713; external cohort had an ACC of 0.668 and AUC of 0.757. The deep learning model yielded an ACC of 0.875 and AUC of 0.907 for the internal, and ACC of 0.833 and AUC of 0.862 for the external cohort. The Radiomics model registered an ACC of 0.819 and AUC of 0.852 internally, and ACC of 0.885 and AUC of 0.903 externally. ResNet-C demonstrated the highest ACC of 0.902 and AUC of 0.934 for the internal, and ACC of 0.885 and AUC of 0.903 for the external cohort.
UNASSIGNED: The ResNet-C model, utilizing bpMRI scanning strategy, accurately assesses bone metastasis (BM) status in newly diagnosed prostate cancer (PCa) patients, facilitating precise treatment planning and improving patient prognoses.