Search

Breathing conditions affect a wide range of people, including those with respiratory issues like asthma and sleep apnea. Smartwatches with photoplethysmogram (PPG) sensors can monitor breathing. However, current methods have limitations due to manual parameter tuning and pre-defined features. To address this challenge, we propose the PPG2RespNet deep-learning framework. It draws inspiration from the UNet and UNet + + models. It uses three publicly available PPG datasets (VORTAL, BIDMC, Capnobase) to autonomously and efficiently extract respiratory signals. The datasets contain PPG data from different groups, such as intensive care unit patients, pediatric patients, and healthy subjects. Unlike conventional U-Net architectures, PPG2RespNet introduces layered skip connections, establishing hierarchical and dense connections for robust signal extraction. The bottleneck layer of the model is also modified to enhance the extraction of latent features. To evaluate PPG2RespNet's performance, we assessed its ability to reconstruct respiratory signals and estimate respiration rates. The model outperformed other models in signal-to-signal synthesis, achieving exceptional Pearson correlation coefficients (PCCs) with ground truth respiratory signals: 0.94 for BIDMC, 0.95 for VORTAL, and 0.96 for Capnobase. With mean absolute errors (MAE) of 0.69, 0.58, and 0.11 for the respective datasets, the model exhibited remarkable precision in estimating respiration rates. We used regression and Bland-Altman plots to analyze the predictions of the model in comparison to the ground truth. PPG2RespNet can thus obtain high-quality respiratory signals non-invasively, making it a valuable tool for calculating respiration rates. [ABSTRACT FROM AUTHOR]

Pulmonary Embolism (PE) has diverse manifestations with different etiologies such as venous thromboembolism, septic embolism, and paradoxical embolism. In this study, a novel attention-based multi-task model is proposed for PE segmentation and detection from Computed Tomography Pulmonary Angiography (CTPA) images. A Y-Net architecture is used to implement this model, which facilitates segmentation and classification jointly, improving performance and efficiency. It is leveraged with Multi Head Attention (MHA), which allows the model to focus on important regions of the image while suppressing irrelevant information, improving the accuracy of the segmentation and detection tasks. The proposed PE-YNet model is tested with two public datasets, achieving a maximum mean detection and segmentation accuracy of 99.89% and 99.83%, respectively, on the CAD-PE challenge dataset. Similarly, it also achieves a detection accuracy of 99.75% and a segmentation accuracy of 99.81% on the FUMPE dataset. Additionally, sensitivity analysis also shows a high sensitivity of 0.9885 for the localization error ɛ = 0 for the CAD-PE dataset, demonstrating the model's robustness against false predictions compared to state-of-the-art models. Further, this model also exhibits lower inference time, size, and memory usage compared to representative models. An automated PE-YNet tool can assist physicians with PE diagnosis, treatment, and prognosis monitoring in the clinical management of CoVID-19. [ABSTRACT FROM AUTHOR]

Optical coherence tomography (OCT) is a widely used imaging technique in ophthalmology for diagnosis and treatment. Recent advances in deep neural networks (DNNs) and vision transformers (ViTs) have paved the way for automated eye/retinal disease classifications and segmentations using OCT or spectral domain OCT (SD-OCT) images. Diabetic macular edema (DME), choroidal neovascularization (CNV), and Drusen are particularly challenging to accurately classify using OCT images because of their subtle differences and intricate features. Currently, the algorithms reported in the literature using DNNs or ViTs are computationally complex, consider fewer diseases, and are less accurate. This study proposes a hybrid SqueezeNet-vision transformer (SViT) model that combines the strengths of SqueezeNet and vision transformer (ViT), capturing local and global features of OCT images to achieve more accurate classification with less computational complexity. The proposed model uses the OCT2017 dataset for training, testing, and validation, and it performs both binary classification (normal vs disorders) as well as multiclass classification (DME, CNV, Drusen, and normal). As compared to state-of-the-art CNN-based and standalone Transformer models, the proposed SViT model achieves an overall classification accuracy of 99.90% for multiclass classification (CNV: 100%, DME: 99.9%, Drusen: 100%, and normal: 100%). With a good generalization ability, the model can be used to improve patient care and clinical decision-making across a broader range of applications. [ABSTRACT FROM AUTHOR]

The early detection of colorectal polyps is crucial for the reduction of mortality rates. However, manually identifying polyps is time-consuming and expensive, increasing the risk of missing them. Our paper aims to address this issue by presenting an automated segmentation approach for colorectal polyps. This paper proposes a method that combines a skip connection with hybrid attention guidance (AG) using attention guidance (AG) and residual path frameworks to identify salient features. Furthermore, we augment test samples using original, horizontal flip, and vertical flip transformations to enhance model robustness through Test Time Augmentation (TTA). The model was trained with Kvasir-seg samples and evaluated on Kvasir-seg and CVC-ClinicDB datasets to gauge generalizability. A significant accuracy (0.9546), a Dice Similarity Coefficient (DSC) of 0.8557, a Cross-section over Union (IoU) of 0.8824, a Recall (0.8221), a Precision (0.8922), an area under Receiver Operating Characteristics (ROC-AUC) of 0.9454, and an area under Precision-Recall (AUC-PR) of 0.8717 were achieved without TTA. Through TTA integration, accuracy (0.9993), DSC (0.8663), IoU (0.8277), Recall (0.8060), Precision (0.9364), and ROC-AUC (0.9587) have been improved. A comparison of our framework with state-of-the-art models demonstrated its effectiveness and segmentation capabilities. Additionally, the proposed model contains only 0.47 million parameters and a weight size of 6.71 MB, illustrating its potential for clinical diagnostics. A computer-aided diagnosis (CAD) system improves patient outcomes by detecting colorectal polyps early and improving segmentation accuracy. [ABSTRACT FROM AUTHOR]

Diabetes mellitus (DM) can cause chronic foot issues and severe infections, including Diabetic Foot Ulcers (DFUs) that heal slowly due to insufficient blood flow. A recurrence of these ulcers can lead to 84% of lower limb amputations and even cause death. High-risk diabetes patients require expensive medications, regular check-ups, and proper personal hygiene to prevent DFUs, which affect 15–25% of diabetics. Accurate diagnosis, appropriate care, and prompt response can prevent amputations and fatalities through early and reliable DFU detection from image analysis. We propose a comprehensive deep learning-based system for detecting DFUs from patients' feet images by reliably localizing ulcer points. Our method utilizes innovative model ensemble techniques—non-maximum suppression (NMS), Soft-NMS, and weighted bounding box fusion (WBF)—to combine predictions from state-of-the-art object detection models. The performances of diverse cutting-edge model architectures used in this study complement each other, leading to more generalized and improved results when combined in an ensemble. Our WBF-based approach combining YOLOv8m and FRCNN-ResNet101 achieves a mean average precision (mAP) score of 86.4% at the IoU threshold of 0.5 on the DFUC2020 dataset, significantly outperforming the former benchmark by 12.4%. We also perform external validation on the IEEE DataPort Diabetic Foot dataset which has demonstrated robust and reliable model performance on the qualitative analysis. In conclusion, our study effectively developed an innovative diabetic foot ulcer (DFU) detection system using an ensemble model of deep neural networks (DNNs). This AI-driven tool serves as an initial screening aid for medical professionals, augmenting the diagnostic process by enhancing sensitivity to potential DFU cases. While recognizing the presence of false positives, our research contributes to improving patient care through the integration of human medical expertise with AI-based solutions in DFU management. [ABSTRACT FROM AUTHOR]

Autism spectrum disorder (ASD) is characterized by divergent etiological factors, comorbidities, severity levels, genetic influences, and functional connectivity (FC) patterns in the brain. In the literature, ASD classification based on age and severity using fMRI data is extremely limited. This study explores the impact of age, symptom severity, and brain FC patterns on the diagnosis of ASD using deep neural networks (DNNs). The ability to classify ASD by age and severity using fMRI data is extremely limited. This study explores the impact of age, symptom severity, and brain FC patterns on the diagnosis of ASD using deep neural networks (DNNs). The FC measures were extracted using Pearson's correlation coefficient (PCC), fractal connectivity (FrC), and nonfractal connectivity (NFrC) from the ABIDE I and II databases. We studied three age groups (6 to 11, 11 to 18, and 6 to 18 years) and two severity groups (ADOS score ≤ 11 and ADOS score > 11). The FC matrices are constructed from blood oxygen level-dependent (BOLD) time series signals, and the heat maps are used to generate features for the convolutional neural network (CNN), MobileNetV2, and DenseNet201 models. The MobileNetV2 classifier achieved 76.25% accuracy, 77.09% sensitivity, and 79.77% precision in the age group of 6 to 11 years using NFrC feature maps compared to other DNNs. According to ADOS total scores above 11, DenseNet201 demonstrated superior performance with 83.45% accuracy, 87.3% sensitivity, and 79.13% precision. Connectivity measured by NFrC consistently outperformed Frc measures. Various combinations of connectivity measures and classifiers consistently showed promising results for the age group of 6–11 years and the severity group with an ADOS score of more than 11. ASD's inherent heterogeneity can be addressed effectively by developing diagnostic models tailored to age and severity. [ABSTRACT FROM AUTHOR]

Hearing loss detection using automated audiometers and artificial intelligence methods has gained increasing attention in recent years. The proposed work aims: (a) to design an automated audiometer to diagnose hearing ability and to evaluate hearing intensity for healthy and profound hearing loss patients within 250 Hz to 8 kHz, (b) to compare the proposed automated audiometer with a conventional audiometer when estimating auditory perception level using pure tone and speech audiometers, and (c) to use the machine learning algorithms to classify hearing loss and normal subjects based on the selected features extracted from speech signals. Participants in the study included 50 healthy individuals and 50 patients with profound hearing loss. In the proposed hardware unit, the transmitted pure-tone signal and the speech signal stimulus are controlled automatically instead of being controlled manually. Using a digital potentiometer, a pure-tone audiometer can be automatically calibrated by varying the frequency and intensity of the generated tones according to the users' responses. During speech audiometric measurements, pre-recorded speech and repeated speech signals are analyzed to estimate speech recognition threshold (SRT) and word recognition score (WRS). The designed audiometer plots the audiogram automatically, estimating SRT and WRS, and classifying the subject as normal or hearing impaired. This study demonstrates the feasibility of using Machine Learning to predict hearing impairment in patients. A support vector machine, a random forest, and an AdaBoost model produced accuracy rates of 98%, 96%, and 96%, respectively, when identifying normal and hearing loss subjects. The proposed audiometer system is miniaturized, portable, and user-friendly in comparison to conventional audiometers. Consequently, the prototype would make it possible for subjects to conduct their own audiometric tests independently and send the results along with their audiogram to a trained medical professional to receive advice. [ABSTRACT FROM AUTHOR]

Lung segmentation algorithms play a significant role in segmenting theinfected regions in the lungs. This work aims to develop a computationally efficient and robust deep learning model for lung segmentation using chest computed tomography (CT) images with DeepLabV3 + networks for two-class (background and lung field) and four-class (ground-glass opacities, background, consolidation, and lung field). In this work, we investigate the performance of the DeepLabV3 + network with five pretrained networks: Xception, ResNet-18, Inception-ResNet-v2, MobileNet-v2 and ResNet-50. A publicly available database for COVID-19 that contains 750 chest CT images and corresponding pixel-labeled images are used to develop the deep learning model. The segmentation performance has been assessed using five performance measures: Intersection of Union (IoU), Weighted IoU, Balance F1 score, pixel accu-racy, and global accuracy. The experimental results of this work confirm that the DeepLabV3 + network with ResNet-18 and a batch size of 8 have a higher performance for two-class segmentation. DeepLabV3 + network coupled with ResNet-50 and a batch size of 16 yielded better results for four-class segmentation compared to other pretrained networks. Besides, the ResNet with a fewer number of layers is highly adequate for developing a more robust lung segmentation network with lesser computational complexity compared to the conventional DeepLabV3 + network with Xception. This present work proposes a unified DeepLabV3 + network to delineate the two and four different regions automatically using CT images for CoVID-19 patients. Our developed automated segmented model can be further developed to be used as a clinical diagnosis system for CoVID-19 as well as assist clinicians in providing an accurate second opinion CoVID-19 diagnosis. [ABSTRACT FROM AUTHOR]

Composite or polymeric insulators have gained significant attention in recent years compared to conventional ceramic disc insulators used in power transmission. The aim of this study is to investigate the performance of ceramic disc and polymeric ceramic 11 kV outdoor products under industrial pollution using laboratory-based pollution performance tests. The insulator performance tests are conducted at various contaminants levels of cement to assess its effectiveness. Statistical analysis of the flashover voltage is investigated through survival functions, hazard functions, and withstand voltages at different failure probability levels. This is an indicator of the level of pollution in the insulator. In this study, normal and Weibull distribution-based statistical models are employed to examine the performance of insulators. The experimental results reported that the ceramic disc insulator and polymeric insulator proved to be more reliable and compatible as insulators in both normal and industrial pollution conditions based on statistical survival functions, hazard functions, and voltage withstand predictions. Based on the statistical analysis of both insulator flashover characteristics and withstand characteristics, the experimental results showed that the normal polymeric insulators and the polluted polymeric insulators have an overall functional reliability performance that is much better than ceramic disc insulators. [ABSTRACT FROM AUTHOR]

Neurotrophic Foot Ulcer (NFU) is most common in patients with diabetes mellitus, and it may result in amputation of the lower extremity (leg and foot). Current methods used for NFU diagnosis are highly complex, report lesser accuracy, and require more computation time and specialists. The main objective of this present work is to design and develop a novel lightweight convolutional neural network (CNN) model to diagnose NFU called NFU-Net. This work utilizes a Diabetic Foot Ulcer (DFU) dataset, and it consists of 1038 abnormal and 641 normal images and naturally transformed technique is used to transfer the data using mathematical operation. We have designed a 22-layers customized CNN with parallel filter architecture to extract highly discriminative features (inter-class and intraclass) from the images to diagnose NFU with two different activation functions, namely Rectified Linear Unit (ReLU) and parametric Rectified Linear Unit (PReLU). The performance of the NFU-Net is compared with the State-Of-The-Art (SOTA) CNN's such as Alex Net, LeNet, DFU-Net, and DFU QUT-Net based on six performance measures (accuracy, precision, sensitivity, specificity, F1 score, Matthews's Correlation Coefficient (MCC) on both original and augmented datasets. The proposed NFU-Net reports an accuracy of nearly 2.5% to 6% higher than conventional CNN's in diagnosing NFU using the DFU dataset. Compared to the traditional CNN's, the proposed network requires lesser network parameters and is computationally efficient. The proposed network could benefit the clinicians for a second opinion about NFU diagnosis. The performance and robustness could be improved while testing the network with other open-source databases. [ABSTRACT FROM AUTHOR]

Stroke is a cerebrovascular disorder, and one of the most common effects of stroke is emotional disturbances. This present work classifies six emotions (anger, sadness, happiness, fear, disgust, and surprise) of two types of stroke (left brain damage and right brain-damage) using electroencephalogram (EEG) signals. EEG signals are collected from 19 each subject of LBD, RBD, and normal control (NC) at a sampling rate of 128 Hz. The IIR Bandpass filter and amplitude thresholding methods are used to reduce noise and artifacts' effects, respectively. Discrete Wavelet Packet Transform is used to extract five EEG frequency bands (alpha, beta, gamma, alpha to gamma, and beta to gamma). A set of nonlinear features are extracted from five different EEG frequency ranges using recurrent quantification analysis. Finally, the extracted features are mapped to six corresponding emotions using three nonlinear classifiers (K nearest neighbor, probabilistic neural network, and random forest). The experimental results indicate that LBD subjects have severe emotional impairment than RBD. The mean of diagonal line length (< L >), recurrence rate, and maximum mean diagonal length (Lmax) feature give maximum classification rate of 85.24% NC, 79.54%, and 79.09% using RF classifier compared to other features. The alpha to gamma (13–49 Hz) band helps identify Stroke emotional state changes compared to other frequency bands. [ABSTRACT FROM AUTHOR]

Sudden cardiac arrest (SCA) prediction using electrocardiogram (ECG) and heart rate variability (HRV) signals has received the attention of researchers in recent years. Ventricular fibrillation (VF) is one of the most common identifiers for SCA. This work aims to investigate the ECG morphological feature, R peak to T-end (R–Tend), to foresee the imminent SCA 5 min before VF onset. ECG signals for SCA and Normal Controls from The MIT-BIH databases are divided into 1-min duration and is used to predict the onset of VF. Four nonlinear features, the Largest Lyapunov Exponent, Hurst exponent, sample entropy, and approximate entropy were extracted from the R–Tend beats and classified using three classifiers: support vector machine, subtractive fuzzy clustering, and neuro-fuzzy classifier. The performance of the proposed methodology confirms that the sample entropy features efficiently predict the SCA five min before VF onset using SVM classifier and produces the maximum mean classification rate of 100% compared to other classifiers. The proposed algorithm using R–Tend beats predicts the onset of SCA better than HRV signals and is computationally efficient. The proposed marker based on ECG morphological characteristics can be used as a tool to predict SCA for smarter healthcare management. [ABSTRACT FROM AUTHOR]

Early-stage detection of lesions is the best possible way to fight breast cancer, a disease with the highest malignancy ratio among women. Though several methods primarily based on deep learning have been proposed for tumor segmentation, it is still a challenging problem due to false positives and the precise boundary detection required for segmentation. In this paper, we propose a Generative Adversarial Network (GAN) based algorithm for segmenting the tumor in Breast Ultrasound images. The GAN model comprises of two modules: generator and discriminator. Residual-Dilated-Attention-Gate-UNet (RDAU-NET) is used as the generator which serves as a segmentation module and a CNN classifier is employed as the discriminator. To stabilize training, Wasserstein GAN (WGAN) algorithm has been used. The proposed hybrid deep learning model is called the WGAN-RDA-UNET. The model is assessed with several quantitative metrics and is also compared with existing methods both quantitatively and qualitatively. The overall Accuracy, PR-AUC, ROC-AUC and F1-score achieved were 0.98, 0.95, 0.89 and 0.88 respectively which are better than most conventional deep net models. The results also showcase the shortcomings of CNN, RDA U-Net and other models and how they can be rectified using the WGAN-RDA-UNET model. [ABSTRACT FROM AUTHOR]

An automated detection system for Parkinson's disease (PD) employing the convolutional neural network (CNN) is proposed in this study. PD is characterized by the gradual degradation of motor function in the brain. Since it is related to the brain abnormality, electroencephalogram (EEG) signals are usually considered for the early diagnosis. In this work, we have used the EEG signals of twenty PD and twenty normal subjects in this study. A thirteen-layer CNN architecture which can overcome the need for the conventional feature representation stages is implemented. The developed model has achieved a promising performance of 88.25% accuracy, 84.71% sensitivity, and 91.77% specificity. The developed classification model is ready to be used on large population before installation of clinical usage. [ABSTRACT FROM AUTHOR]

The objective of this research was to investigate the relationship between emotion recognition and lateralization of motor onset in Parkinson's disease (PD) patients using electroencephalogram (EEG) signals. The subject pool consisted of twenty PD patients [ten with predominantly left-sided (LPD) and ten with predominantly right-sided (RPD) motor symptoms] and 20 healthy controls (HC) that were matched for age and gender. Multimodal stimuli were used to evoke simple emotions, such as happiness, sadness, fear, anger, surprise, and disgust. Artifact-free emotion EEG signals were processed using the auto regressive spectral method and then subjected to repeated ANOVA measures. No group differences were observed across behavioral measures; however, a significant reduction in EEG spectral power was observed at alpha, beta and gamma frequency oscillations in LPD, compared to RPD and HC participants, suggesting that LPD patients (inferred right-hemisphere pathology) are impaired compared to RPD patients in emotional processing. We also found that PD-related emotional processing deficits may be selective to the perception of negative emotions. Previous findings have suggested a hemispheric effect on emotion processing that could be related to emotional response impairment in a subgroup of PD patients. This study may help in clinical practice to uncover potential neurophysiologic abnormalities of emotional changes with respect to PD patient's motor onset. [ABSTRACT FROM AUTHOR]

Recent studies show right hemisphere has a unique contribution to emotion processing. The present study investigated EEG using non-linear measures during emotional processing in PD patients with respect to motor symptom asymmetry (i.e., most affected body side). We recorded 14-channel wireless EEGs from 20 PD patients and 10 healthy age-matched controls (HC) by eliciting emotions such as happiness, sadness, fear, anger, surprise and disgust. PD patients were divided into two groups, based on most affected body side and unilateral motor symptom severity: left side-affected (LPD, n = 10) or right side-affected PD patients (RPD, n = 10). Nonlinear analysis of these emotional EEGs were performed by using approximate entropy, correlation dimension, detrended fluctuation analysis, fractal dimension, higher order spectra, hurst exponent (HE), largest Lyapunov exponent and sample entropy. The extracted features were ranked using analysis of variance based on F value. The ranked features were then fed into classifiers namely fuzzy K-nearest neighbor and support vector machine to obtain optimal performance using minimum number of features. From the experimental results, we found that (a) classification performance across all frequency bands performed well in recognizing emotional states of LPD, RPD, and HC; (b) the emotion-specific features were mainly related to higher frequency bands; and (c) predominantly LPD patients (inferred right-hemisphere pathology) were more impaired in emotion processing compared to RPD, as showed by a poorer classification performance. The results suggest that asymmetric neuronal degeneration in PD patients may contribute to the impairment of emotional communication. [ABSTRACT FROM AUTHOR]

Parkinson's disease (PD) is not only characterized by its prominent motor symptoms but also associated with disturbances in cognitive and emotional functioning. The objective of the present study was to investigate the influence of emotion processing on inter-hemispheric electroencephalography (EEG) coherence in PD. Multimodal emotional stimuli (happiness, sadness, fear, anger, surprise, and disgust) were presented to 20 PD patients and 30 age-, education level-, and gender-matched healthy controls (HC) while EEG was recorded. Inter-hemispheric coherence was computed from seven homologous EEG electrode pairs (AF3-AF4, F7-F8, F3-F4, FC5-FC6, T7-T8, P7-P8, and O1-O2) for delta, theta, alpha, beta, and gamma frequency bands. In addition, subjective ratings were obtained for a representative of emotional stimuli. Interhemispherically, PD patients showed significantly lower coherence in theta, alpha, beta, and gamma frequency bands than HC during emotion processing. No significant changes were found in the delta frequency band coherence. We also found that PD patients were more impaired in recognizing negative emotions (sadness, fear, anger, and disgust) than relatively positive emotions (happiness and surprise). Behaviorally, PD patients did not show impairment in emotion recognition as measured by subjective ratings. These findings suggest that PD patients may have an impairment of inter-hemispheric functional connectivity (i.e., a decline in cortical connectivity) during emotion processing. This study may increase the awareness of EEG emotional response studies in clinical practice to uncover potential neurophysiologic abnormalities. [ABSTRACT FROM AUTHOR]

A stress assessment based on the electrocardiography (ECG) and heart rate variability (HRV) signals is described in this paper. The Stroop color word test (stressor) was used to induce stress, and the ECG signal was acquired throughout the experiment to identify the variations that are induced by this stressor. A total of 10 female subjects (aged 20-25 years) participated in this study. A time and frequency domain analysis of the HRV and ECG signals was done to extract the stress-related features. A total of five frequency bands and ratios of the HRV signal were used to analyze the new and existing statistical features. The results indicate that significant changes between the normal and stressed states are more evident with a classification accuracy of 79.17 %. Alternatively, the low frequency range (0.04-0.5 Hz) of the ECG signal (0-100 Hz) was used to identify the effect of stress instead of the usual frequency domain analysis of the HRV signal (0.04-0.5 Hz). To extract the stress-related features of the ECG signal, a discrete wavelet transform based feature extraction was performed using the ' db4' and ' coif5' wavelet functions. A set of eight statistical features was extracted from the two different frequency bands and the three frequency band ratios. All of the extracted features were classified into two states (stress and normal) using the simple non-linear K-nearest neighbor classifier. The experimental results gave the maximum average accuracy of 94.58 and 94.22 % with the ' db4' and ' coif5' wavelet functions, respectively. Remarkably, the classification results obtained with the features of the ECG and HRV signals were completely independent of the post-task questionnaire. The outcome of this work was helpful to develop the multiple physiological signal based stress system using optimal features in these two signals. [ABSTRACT FROM AUTHOR]