Long acquisition times, a staple characteristic of Magnetic Resonance Imaging (MRI) can be shortened by aggressively undersampling k-space and solving the resulting ill-posed reconstruction problem using learning based sparsity priors. Unrolled optimization networks such as FISTA-Net provide an interpretable framework with strong data consistency but struggle to restore fine textures at very low sampling ratios ( ≤ 10 ≤ 5 % brain data from the CC-359 benchmarked dataset, the proposed model achieves 39.5 dB PSNR and 0.9142 SSIM, surpassing state-of-the-art unrolled baselines (FISTA-Net) and GAN (HARA-GAN, RSCA-GAN) by up to 2.6 dB and 0.05 SSIM, respectively. The results demonstrate that the integration of interpretability and adversarial detail enhancement yields fast, high-fidelity reconstructions suitable for time-critical clinical workflows.

The spectral lines in the low-frequency analysis and recording (LOFAR) spectrum are the primary sources of in formation that underwater passive sonar detection systems rely on. The quality of the spectrum degrades because of the marine environment. Conventional adaptive line enhancers (ALEs) fail to incorporate simultaneous adapt ability to time and frequency variations, limiting their effectiveness in varying conditions. This paper proposes an L1-norm based wavelet domain ALE that utilizes the multi-resolution and localized time-frequency analysis capabilities of the sliding discrete wavelet transform (SDWT) to effectively isolate spectral components. By in tegrating an iteratively weighted L1-norm regularization term into the adaptive weight update mechanism, the proposed approach achieves efficient noise suppression. The experimental findings demonstrate that when tested on the ShipsEar dataset, Deepship dataset, and synthetic data, the proposed method achieves better visual quality of the enhanced LOFAR spectrum with a sharp spectral line among all compared methods. The output signal-tonoise ratio (SNR) gain of the proposed method is 3.57 dB higher than that of the state-of-the-art method, namely, sparsity ALE based on mixed weighted error (MWE) and Shannon entropy (SE) criteria, and 6.72 dB higher than sparsity based ALE when evaluated on simulated data under an extreme noise condition with SNR-25 dB.

Reconstructing high quality magnetic resonance (MR) images from compressed sensing (CS) data remains a difficult problem, especially when only a small portion of k-space is sampled. In this work, we introduce Graph-FISTA-Net, an interpretable unrolled network designed for fast CS-MRI reconstruction. Instead of following the usual fixed iteration order in FISTA based models, our model learns a directed acyclic graph that connects different stages of the optimization process. Each node captures how information is shared between iterations through a temperature-controlled adjacency matrix, enabling the network to discover its own data-driven momentum pattern. Entropy and sparsity regularizations are incorporated to maintain stability and ensure monotone convergence. Experiments with T1-weighted brain magnetic resonance data ($10-40 \%$ sampling) show consistent improvements in PSNR and SSIM over FISTA-Net without increasing inference time. Visualizations of the learned graph reveal clear and interpretable update behaviors during reconstruction.

Artificial intelligence (AI) predictions are widely used to address challenges in the heart health sector, such as providing clinical decision support. Early detection of valvular heart disease (VHD) is effective in reducing critical cardiac problems and sudden death. This review proposes investigating methods for automatically diagnosing heart disease from phonocardiogram (PCG) signals using various advanced Machine Learning (ML) and Deep Learning (DL) models. This study also aimed to provide an overview of ongoing research on PCG signal processing and to pinpoint areas that warrant additional investigation. Several Scopus-indexed research forums, such as IEEE, Science Direct, Frontiers, MDPI, and Computing in Cardiology, as well as databases such as the Michigan Heart Sound Library (MHSL), Github, and Physio-Net on the classification of AI-related PCG signals, were considered to construct this review with 199 relevant research articles covering the period from 2016 to 2024. The early diagnosis and prediction of heart valve disease are the domains in which machine learning and deep learning models were most commonly used. The performance of earlier detection has increased significantly according to advanced techniques of PCG signal classification. A limited number of studies have compared and analyzed categorization measures such as F score, sensitivity, accuracy, precision, and specificity. However, a mean increase in the predicted accuracy was observed, depending on the various advanced techniques and classifiers used.

Analysis of the 12-lead electrocardiogram (ECG) is very crucial for cardiac anomaly detection. Acquiring 12-lead ECG without a clinical setting is very challenging. Previous studies have extensively looked into creating a 12-lead ECG from a smaller lead set to enhance patient comfort and simplify ambulatory monitoring. This research presents a unique deep-learning model customized for synthesizing 12-lead ECGs in a patient-specific manner The proposed method utilizes the increased inter-lead correlation in the ECG signal by employing extended long-short-term memory (xLSTM) networks along with Bayesian optimization methods, which are known to be effective for analyzing sequential data and aim to enhance the quality of reconstruction. Different diagnostic criteria and similarity metrics are used to evaluate the results of the study. The suggested structure is proven, and successful reconstruction is possible as it can detect important clinical characteristics and provide a robust defence against interference. In the case of the proposed method, the normalized mean square error is very low compared to the existing state-of-the-art methods, and the correlation coefficient is also more than 99%

Machine learning (ML) has become a popular technique for various automation tasks in the era of Industry 4.0, such as the analysis and synthesis of visual data such as images and videos, natural language and speech, financial data, and biomedical applications. However, ML-based automation techniques are facing difficulties like decision-making, thus incorporating user expertise into the system might be advantageous. The goal of adding human domain expertise with ML-based automation is to provide more accurate prediction models. Human-in-the-loop (HITL) systems that integrate human expertise with ML algorithms are becoming more and more common in various industries. However, there are a number of methodological, technical, and ethical difficulties with the development and application of HITL systems. This paper aims to explore the methodologies, challenges, and opportunities associated with HITL systems implementations.We also discuss a number of issues that must be resolved for HITL systems to be effective, including data quality, bias, and user engagement. Besides, we also explored several approaches that can be utilized to enhance the performance of HITL systems, such as active learning (AL), iterative ML, and reinforcement learning, as well as the current state of the art in HITL systems.We also selectively highlighted the advantages of HITL systems, such as their potential to increase decision-making process accountability and transparency by utilizing human experience to improve ML decision-making capability. The paper will be very useful for researchers, practitioners, and policymakers.

PurposeThere are a number of algorithms for smooth l0-norm (SL0) approximation. In most of the cases, sparsity level of the reconstructed signal is controlled by using a decreasing sequence of the modulation parameter values. However, predefined decreasing sequences of the modulation parameter values cannot produce optimal sparsity or best reconstruction performance, because the best choice of the parameter values is often data-dependent and dynamically changes in each iteration.ApproachWe propose an adaptive compressed sensing magnetic resonance image reconstruction using the SL0 approximation method. The SL0 approach typically involves one-step gradient descent of the SL0 approximating function parameterized with a modulation parameter, followed by a projection step onto the feasible solution set. Since the best choice of the parameter values is often data-dependent and dynamically changes in each iteration, it is preferable to adaptively control the rate of decrease of the parameter values. In order to achieve this, we solve two subproblems in an alternating manner. One is a sparse regularization-based subproblem, which is solved with a precomputed value of the parameter, and the second subproblem is the estimation of the parameter itself using a root finding technique.ResultsThe advantage of this approach in terms of speed and accuracy is illustrated using a compressed sensing magnetic resonance image reconstruction problem and compared with constant scale factor continuation based SL0-norm and adaptive continuation based l1-norm minimization approaches. The proposed adaptive estimation is found to be at least twofold faster than automated parameter estimation based iterative shrinkage-thresholding algorithm in terms of CPU time, on an average improvement of reconstruction performance 15% in terms of normalized mean squared error.ConclusionsAn adaptive continuation-based SL0 algorithm is presented, with a potential application to compressed sensing (CS)-based MR image reconstruction. It is a data-dependent adaptive continuation method and eliminates the problem of searching for appropriate constant scale factor values to be used in the CS reconstruction of different types of MRI data.

Synthesis of a 12-lead electrocardiogram from a reduced lead set has previously been extensively studied in order to meet patient comfort, minimise complexity, and enable telemonitoring. Traditional methods relied solely on the inter-lead correlation between the standard twelve leads for learning the models. The 12-lead ECG possesses not only inter-lead correlation but also intra-lead correlation. Learning a model that can exploit this spatio-temporal information in the ECG could generate lead signals while preserving important diagnostic information. The proposed approach takes leverage of the enhanced inter-lead correlation of the ECG signal in the wavelet domain. Long-short-term memory (LSTM) networks, which have emerged as a powerful tool for sequential data mining, are a type of recurrent neural network architecture with an inherent capability to capture the spatiotemporal information of the heart signal. This work proposes the deep learning architecture that utilizes the discrete wavelet transform and the LSTM to reconstruct a generic 12-lead ECG from a reduced lead set. The experimental results are evaluated using different diagnostic measures and similarity metrics. The proposed framework is well founded, and accurate reconstruction is possible as it can capture clinically significant features and provides a robust solution against noise.

Conventional machine learning approaches have shown great promise in video action recognition tasks. However, these approaches have a fatal flaw: where in they fail in dynamically adapting to new stimuli in an ever evolving environment without losing any of it's previously learned knowledge. This missing ability in general machine learning is called Continual learning, a key technique in attaining Artificial General Intelligence systems (AGI's) making them capable of adapting to real-world scenario's. Catastrophic forgetting is the obstacle that hinders machine learning models from achieving continual learning. There are various approaches for embedding continual learning ability in a standard neural network architecture. In computer vision, even though continual learning has been explored in tasks like image classification, object detection and video recognition, most studies explore a replay-based continual learning approach that has a computational overhead of having to store samples from previous tasks. Here, we propose a continual learning video action recognition architecture based on a Parameter Allocation approach. The proposed method can continually learn video recognition tasks without losing its previously acquired knowledge in comparison to the traditional models and is also free from storing exemplars from previous tasks.

Artificial intelligence (AI) systems are trained to solve complex problems and learn to perform specific tasks by using large volumes of data, such as prediction, classification, recognition, decision-making, etc. In the past three decades, AI research has focused mostly on the model-centric approach compared to the data-centric approach. In the model-centric approach, the focus is to improve the code or model architecture to enhance performance, whereas in data-centric AI, the focus is to improve the dataset to enhance performance. Data is food for AI. As a result, there has been a recent push in the AI community toward data-centric AI from model-centric AI. This paper provides a comprehensive and critical analysis of the current state of research in data-centric AI, presenting insights into the latest developments in this rapidly evolving field. By emphasizing the importance of data in AI, the paper identifies the key challenges and opportunities that must be addressed to improve the effectiveness of AI systems. Finally, this paper gives some recommendations for research opportunities in data-centric AI.

This study presents a unique method that can be used for obtaining a 12-lead ECG from a simplified ECG lead set using deep neural network-based patient-specific nonlinear reconstruction techniques. Conventionally, ten electrodes are used to acquire a full 12-lead ECG, which introduces significant challenges in terms of device portability, patient comfort, and operational costs. By synthesizing a full ECG from a smaller set of leads, it can make ECG more feasible for continuous and remote wearable health monitoring. In this study, we used 3-input leads and 5-target leads for deep neural network training. We reconstructed 5-target leads using only 3-input leads after the model was trained. We trained and tested the model using LSTM on the PTB diagnostic ECG database and assess the correspondence between the original and reconstructed signals using statistical analysis. The LSTM-based approach has proven to be sufficiently accurate. The proposed method gives an average accuracy of 0.9927 and can be deployed in remote wearable healthcare applications. The reconstructed signal is compared with the existing state-of-the-art method. Here the proposed method gives NMSE values lower than the polynomial lasso method.

Energy consumption involved in wireless transmission poses a major challenge in the implementation of wireless body area networks (WBAN). Compressed sensing (CS)-based multichannel electrocardiogram (ECG) compression is a new paradigm in signal acquisition and reconstruction; a viable alternative for traditional wavelet-based signal reconstruction. However, several challenges must be addressed to achieve efficient and reliable ECG compression in real-world wireless healthcare systems. This paper presents a comprehensive review focused on the evolution of compressed sensing-based energy-efficient single-channel (S-) and multi-channel (M-) ECG data compression techniques. It is observed that the performance of different compression techniques depends on several diagnostic or non-diagnostic test parameters. The present study could be useful for researchers to analyze the state-of-the-art compression techniques in e-healthcare applications. We discuss the challenges associated with implementing ECG compression in wireless healthcare systems, such as signal quality, interoperability, and privacy concerns. We also explore the potential future directions for research in this area, including the development of novel algorithms for compressed sensing-based ECG compression, the integration of artificial intelligence and deep learning techniques, and the exploration of new application areas for wireless healthcare systems. This paper will serve as a good reference for the researcher interested in the area of wireless transmission for WBAN applications.

Magnetic resonance imaging (MRI) is a vital nonionizing medical imaging modality, but its widespread clinical utility is hindered by the long data acquisition times. Compressed sensing (CS) has emerged as a promising solution to this problem, as it allows for significant scan time reduction without compromising image quality. However, CS introduces the challenge of solving ill-posed inverse problems, which are typically addressed using first-order iterative proximal gradient methods. These traditional methods, while effective, rely on fixed tuning parameters that can limit their ability to reconstruct high-quality images, especially in cases of extreme undersampling. In this paper, we propose FISTA-Net, an unrolled neural network architecture that mimics the FISTA algorithm but replaces its fixed parameters with learnable counterparts. By optimizing these parameters through training, FISTA-Net significantly en-hances the quality of reconstructed images. Experimental results demonstrate that at different undersampling ratios, FISTA-Net achieves reconstruction with an average PSNR of 34.81dB and an SSIM of 0.9198, compared to 31.59dB and 0.7612 for traditional FISTA. These findings underscore the superiority of FISTA-Net, making it a more effective approach for CS- MRI reconstruction.

Data-centric artificial intelligence (AI) (DCAI) has the potential to bring significant benefits to society; however, it also poses significant challenges and potential risks. It is crucial to approach the development and deployment of DCAI systems with caution, taking into account the potential societal impacts and working to mitigate any negative effects. DCAI technology is now an essential part of operations for many of the world's largest software and hardware industries. These industries offer a range of AI and machine-learning (ML) services, tools, and platforms to society to help businesses process and analyze data. By leveraging data, these industries are able to drive innovation, optimize their operations, and gain a competitive advantage in the market. From personalized recommendations to optimized manufacturing processes, data analytics and ML algorithms are being used to improve the overall customer experience, increase efficiency, and identify new opportunities for growth. As data continue to play an increasingly important role in business operations, it is likely that more companies will adopt these technologies to stay ahead of the curve and succeed in today's data-driven world [1] , [2] , [3] .

Sparse representation-based single image super-resolution (SISR) methods use a coupled overcomplete dictionary trained from high-resolution images/image patches. Since remote sensing (RS) satellites capture images of large areas, these images usually have poor spatial resolution and obtaining an effective dictionary as such would be very challenging. Moreover, traditional patch-based sparse representation models for reconstruction tend to give unstable sparse solution and produce visual artefact in the recovered images. To mitigate these problems, in this article, we have proposed an adaptive joint sparse representation-based SISR method that is dependent only on the input low-resolution image for dictionary training and sparse reconstruction. The new model combines patch-based local sparsity and group sparse representation-based nonlocal sparsity in a single framework, which helps in stabilizing the sparse solution and improve the SISR results. The experimental results are evaluated both visually and quantitatively for several RGB and multispectral RS datasets, where the proposed method shows improvements in peak signal-to-noise ratio by 1–4 dB and 2–3 dB over the state-of-the-art sparse representation- and deep learning-based SR methods, respectively. Land cover classification applied on the super-resolved images further validate the advantages of the proposed method. Finally, for practical RS applications, we have performed parallel implementation in general purpose graphics processing units and achieved significant speed ups (30–40×) in the execution time.

This paper presents a patient-specific approach for reconstructing the standard 12-lead ECG from a minimal lead set. The 12-lead ECG signal acquisition impediment using ten electrodes comprises ambulatory monitoring, personalized healthcare, remote healthcare, and pediatric ECG lead placement. Furthermore, concurrently processing signals from multiple electrodes enhances the intricacy as well as the cost. Synthesizing 12-lead ECG from a reduced lead set becomes a better solution. This study proposes a recurrent neural network (RNN) long short-term memory (LSTM) to synthesize standard 12-lead ECG from the three predictor leads. The four performance metrics, namely correlation coefficient (cc), root mean square error (RMSE), and wavelet energy diagnostic distortion (WEDD), are employed to evaluate the performance of the proposed method. The proposed model obtained fine reconstruction quality and achieved better performance than most of the previously established works without compromising diagnostic information.

Background: Brain imaging techniques provide the ability to noninvasively map the structure and functions of the brain. Brain vascular malformation mainly affects people in the 5th decade followed by 4th and 3rd decade. Aims and Objectives: The aim of the study was to compare the diagnostic supremacy of computed tomography angiography (CTA)/magnetic resonance angiography (MRA) and Digital subtraction angiography (DSA) for the detection of intracranial vascular anomalies. Materials and Methods: An observational descriptive study with cross-sectional design was performed among 50 patients of both sexes undergoing DSA test at Medical College and Hospital, Kolkata with acute stroke syndrome or any other symptoms suggesting intracranial vascular lesion, who were investigated with one or more index tests and a reference standard diagnosed by computed tomography (CT) or MR scanning or other parameters. DSA served as the standard of reference for presence of intracranial vascular anomalies. Results: Out of the 50 patients included in the study, 41 were diagnosed with vascular malformations by DSA. Moya moya disease was diagnosed in three and distal AV fistula in six patients. In the 41 patients with vascular malformations, CTA could correctly identify 17 (41.5%) cases whereas MRA could identify 73.2% cases. Conclusion: DSA can be used for both diagnostic and interventional angiography. Its high spatial and temporal resolution have maintained DSA as a very important tool. The study reveal DSA is more superior to accurate angioarchitectural delineation of different intracerebral vascular malformation.

In clinical practice, continuous recording and monitoring of the standard 12-lead electrocardiogram (ECG) is often not feasible. The emerging technology and advancement to record the ECG signal without the help of the medical expert’s in-home care or ambulatory conditions with minimal complexity have become more common in recent times. We aim to devise a model to obtain the 12-lead ECG from a reduced number of leads to reduce the intricacy and enhance patient comfort and care. We propose a discrete wavelet transform (DWT) based artificial neural network (ANN) model that transforms a 3-lead ECG into a standard 12-lead ECG without losing diagnostic information. Prominent distortion measures, namely, correlation coefficient, R2 statistics, and wavelet energy diagnostic distortion (WEDD) are employed to evaluate the quality of the synthesis by the proposed model. The performance of the suggested model is compared with the antecedent models. The experimental result shows that the proposed technique can successfully synthesize the standard 12-lead ECG from the reduced lead sets.

Multi-channel electrocardiogram (MECG) compression on lightweight wireless body area network (WBAN) is highly challenging for long-term eHealthcare monitoring. Energy consumption involved in wireless transmission poses an obstacle in the implementation of WBAN. MECG is widely used to diagnose cardiovascular diseases (CVDs), which require a considerable time to extract sufficient clinically relevant data from subjects. Since wireless MECG data transmission dominates energy cost and memory consumption of wireless sensor nodes, it is favorable to reduce the data size without any loss of diagnostic information. Compressed sensing (CS) provides efficient encoding schemes for data reduction and energy consumption in wireless transmission. We propose an adaptive multiple dictionary learning-based joint CS model for MECG compression, which exploits spatial correlation and adaptive features existing in MECG signals. We demonstrate performance of the proposed algorithm with the Physikalisch-Technische Bundesanstalt database (PTB) for MECG compression. Results from the proposed scheme can achieve excellent reconstruction quality with fewer measurements than other existing CS-based approaches. It is the most suitable technique for long-term MECG compression using lightweight WBAN devices.