The damage to the heart valves causes heart valve disorders (HVDs). The detection of HVDs is crucial in a clinical study as these diseases may cause congestive heart failure, hypertrophy, and stroke. The phonocardiogram (PCG) signal reveals information regarding the mechanical activity of the heart. The early detection of HVDs using PCG signal is vital to minimize the chances of cardiac arrest and other cardiac complications. This article proposes the time–frequency-domain deep learning (TFDDL) framework for automatic detection of HVDs using PCG signals. The time–frequency (TF)-domain representations of PCG signals are evaluated using both time-domain polynomial chirplet transform (TDPCT) and frequency-domain polynomial chirplet transform (FDPCT). The deep convolutional neural network (CNN) model is used to detect four types of HVDs using the TF images of PCG signals obtained using both the TDPCT and FDPCT methods. The proposed TFDDL approach is evaluated using PCG signals from public databases. For the detection of HVDs using TDPCT- and FDPCT-based TF images of PCG signals, the suggested approach has achieved overall accuracy values of 99% and 99.48%, respectively. For the classification of normal and abnormal heart sound classes, the proposed TFDDL approach has obtained an accuracy of 85.16% using PCG signals from the Physionet challenge 2016 database. The proposed TFDDL framework is compared with TF-domain transfer learning models such as residual network (ResNet-50) and visual geometry group (VGGNet-16). The overall accuracy values obtained using VGGNet-16 and ResNet-50 are less than the proposed deep CNN model for the detection of HVDs. The proposed TFDDL model can be validated in real-time using heart sound signals recorded from different subjects for automated identification of HVDs.

In this letter, a promising method is proposed to automatically detect pulmonary diseases (PDs) from lung sound (LS) signals. The modes of LS signal is evaluated using empirical wavelet transform (EWT) with fixed boundary points (FBPs). The time-domain (Shannon entropy) and frequency domain (peak amplitude and peak frequency) features have been extracted from each mode. The classifiers such as support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and light gradient boosting machine (LGBM) have been chosen to detect PDs using features of LS signals automatically. The performance of the proposed method has been evaluated using LS signals obtained from a publicly available database. The detection accuracy values such as 80.35%, 83.27%, 99.34%, and 77.13% have been obtained using the LGBM classifier with 5-fold cross-validation (CV) for normal versus asthma, normal versus pneumonia, normal versus chronic obstructive PD (COPD), and normal versus pneumonia versus asthma versus COPD classification schemes. For the normal versus pneumonia versus asthma classification scheme, the proposed method has achieved an accuracy value of 84.76% which is higher than the existing approaches using LS signals.

Myocardial infarction (MI) is a life-debilitating emergency in which there is a lack of blood flow in the heart muscle, resulting in permanent damage to the myocardium and sudden cardiac death. The 12-lead electrocardiogram (ECG) is a standardized diagnostic test conducted in hospitals to detect and localize MI-based heart disease. To diagnose MI, the cardiologist visualizes the alternations in the patterns of the 12-lead-based ECG trace image. The automated detection of MI from the 12-lead-based ECG trace image using artificial intelligence (AI) based approaches is important in the clinical study for the accurate diagnosis of MI disease. This paper proposes a novel eigendomain-based deep representation learning approach to automatically detect MI using 12-lead ECG trace images. The singular value decomposition (SVD) and eigendomain grouping are employed to evaluate five modes or components from the 12-lead ECG trace image. The EfficientNetV2B2-based transfer learning model extracts feature maps from the 12-lead ECG trace image and all five modes. The global average pooling (GAP), batch normalization (BN), drop-out, and soft-max layers are used for each feature map to obtain the probability scores. The concatenated probability scores of all feature maps, followed by the dense layer and output layer, are used to detect MI. A public database containing the 12-lead ECG trace images is used to evaluate the performance of the proposed approach. The results show that for the MI class, the proposed approach has achieved the accuracy value of 100%. Similarly, for normal versus MI versus other cardiac arrhythmia-based disease classification schemes, the proposed approach has obtained the overall accuracy, F1-score, specificity, and sensitivity values of 99.03%, 99.01%, 99.49%, and 98.96%, respectively using 5-fold cross-validation (CV). The suggested approach has demonstrated higher overall accuracy than twenty-four existing transfer learning-based models to detect MI using 12-lead ECG trace images.

In this letter, we propose an automated approach for recognizing imagined commands using multi-channel electroencephalogram (MEEG) signals. The multivariate fast and adaptive empirical mode decomposition (MFAEMD) method decomposes the MEEG signals into modes. The slope domain entropy (SDE) and $L_1$ -norm features are evaluated from the modes of MEEG signals. The machine learning models such as k-nearest neighbor (KNN), sparse representation classifier (SRC), and dictionary learning (DL) are used for the imagined command classification tasks. The approach is evaluated using MEEG signals from a public database. The proposed approach has obtained average accuracy values of 60.72%, 59.73%, and 58.78% using a dictionary learning model and selected features for left versus right, up versus down, forward versus backward based imagined command categorization tasks.

This letter proposes the multiscale domain gradient boosting (MDGB) based approach for the automated recognition of imagined vowels using the multichannel electroencephalogram (MCEEG) signals. The multiscale analysis of the MCEEG signals is performed using multivariate automatic singular spectrum analysis (MASSA) and multivariate fast and adaptive empirical mode decomposition (MFAEMD) methods. The features such as bubble entropy, energy, slope domain entropy (SDE), sample entropy, and L1-norm are evaluated from the multiscale domain modes of the MCEEG signals. The extreme gradient boosting (XGBoost) and light gradient boosting machine (LGBM) models are employed for imagined vowel recognition task as //a// versus //e// versus //i// versus //o// versus //u// using the multiscale domain features of the MCEEG signals. A publicly available MCEEG database has been used to test the performance of the proposed approach. The results demonstrate that the proposed approach has achieved an overall accuracy (OA) of 51.47% which is higher as compared to other imagined vowel recognition methods using the same database comprising of the MCEEG signals.

The automated detection of tuberculosis (TB) and pneumonia (PN) from chest X-ray (CXR) images using artificial intelligence (AI) is challenging in clinical studies for rapid diagnosis and initiation of treatment. This letter proposes the field programmable gate array (FPGA)-based hardware implementation of a novel lightweight deep convolutional neural network (DCNN) model to detect PN and TB ailments using CXR images. Initially, the proposed DCNN (consisting of 10 layers) is trained using the Google Cloud central processing unit (CPU) to obtain the model weight and bias parameters. Then, the register transfer logic (RTL) for the trained DCNN model is generated by the VIVADO high-level synthesis (HLS) framework using HLS for machine learning (HLS4ML) with fixed point representation (8-bit for integer and 12-bit for the fractional part). The hardware implementation of the suggested DCNN model is performed using the PYNQ-Z2 FPGA framework to detect TB and PN diseases automatically. The experimental results demonstrate that the proposed DCNN model has obtained accuracy values of 96.39% and 95.63% on the Google-Cloud CPU and PYNQ-Z2 FPGA frameworks using 422 CXR images in the inference phases. The inference time of the proposed DCNN model on the PYNQ-Z2 FPGA framework is reduced by 85.19% compared to the CPU-based implementation. The suggested DCNN model has only 1831 parameters, less than the transfer learning (TFL) and existing CNN-based models to detect TB and PN using CXR images.