BACKGROUND:Digitization of paper-based electrocardiograms (ECGs) enables long-term preservation, fast transmission, and advanced analysis. Traditional methods for digitizing ECGs face significant challenges, particularly in real-world scenarios with varying image quality. State-of-the-art solutions often require manual input and are limited by their dependence on high-quality scans and standardized layouts. METHODS:This study introduces a fully automated, deep learning-based approach for high precision ECG digitization. In the normalization phase, a standardized grid structure is detected, and image distortions are corrected. Next, the reconstruction phase uses deep learning techniques to extract and digitize the leads, followed by post-processing to refine the signal. This approach was evaluated using the publicly available PMcardio ECG Image Database (PM-ECG-ID), comprising 6000 ECG images reflecting diverse real-world scenarios and smartphone-based image acquisitions. Performance was assessed using Pearson's correlation coefficient (PCC), root mean squared error (RMSE), and signal-to-noise ratio (SNR). RESULTS:The ECG digitization solution demonstrated an average PCC consistently exceeding 0.91 across all leads, SNR above 12.5 dB and RMSE below 0.10 mV. The time to ECG digitization was consistently less than 7 s. The average failure rate was 6.62 % across leads, with most failures occurring under extreme conditions such as severe blurring or significant image degradation. The solution maintained robust performance even under challenging scenarios, such as low-resolution images, distorted grids, and overlapping signals. CONCLUSION:Our deep learning-based approach for ECG digitization delivers high-precision signals, effectively addressing real-world challenges. This fully automated method enhances the accessibility and utility of ECG data by enabling convenient digitization via smartphones, unlocking advanced AI-driven analysis.

Aims A majority of acute coronary syndromes (ACS) present without typical ST elevation. One-third of non-ST-elevation myocardial infarction (NSTEMI) patients have an acutely occluded culprit coronary artery [occlusion myocardial infarction (OMI)], leading to poor outcomes due to delayed identification and invasive management. In this study, we sought to develop a versatile artificial intelligence (AI) model detecting acute OMI on single-standard 12-lead electrocardiograms (ECGs) and compare its performance with existing state-of-the-art diagnostic criteria. Methods and results An AI model was developed using 18 616 ECGs from 10 543 patients with suspected ACS from an international database with clinically validated outcomes. The model was evaluated in an international cohort and compared with STEMI criteria and ECG experts in detecting OMI. The primary outcome of OMI was an acutely occluded or flow-limiting culprit artery requiring emergent revascularization. In the overall test set of 3254 ECGs from 2222 patients (age 62 +/- 14 years, 67% males, 21.6% OMI), the AI model achieved an area under the curve of 0.938 [95% confidence interval (CI): 0.924-0.951] in identifying the primary OMI outcome, with superior performance [accuracy 90.9% (95% CI: 89.7-92.0), sensitivity 80.6% (95% CI: 76.8-84.0), and specificity 93.7 (95% CI: 92.6-94.8)] compared with STEMI criteria [accuracy 83.6% (95% CI: 82.1-85.1), sensitivity 32.5% (95% CI: 28.4-36.6), and specificity 97.7% (95% CI: 97.0-98.3)] and with similar performance compared with ECG experts [accuracy 90.8% (95% CI: 89.5-91.9), sensitivity 73.0% (95% CI: 68.7-77.0), and specificity 95.7% (95% CI: 94.7-96.6)]. Conclusion The present novel ECG AI model demonstrates superior accuracy to detect acute OMI when compared with STEMI criteria. This suggests its potential to improve ACS triage, ensuring appropriate and timely referral for immediate revascularization.

Pruning large language models (LLMs) is a challenging task due to their enormous size. The primary difficulty is fine-tuning the model after pruning, which is needed to recover the lost performance caused by dropping weights. Recent approaches have either ignored fine-tuning entirely, focusing on efficient pruning criteria, or attempted layer-wise weight updates, preserving the behavior of each layer. However, even layer-wise weight updates can be costly for LLMs, and previous works have resorted to various approximations. In our paper, we propose a fast and optimal weight update algorithm for pruned layers based on the Alternating Direction Method of Multipliers (ADMM). Coupled with a simple iterative pruning mask selection, our algorithm achieves state-of-the-art pruning performance across a wide range of LLMs. Code is available at https://github.com/fmfi-compbio/admm-pruning.

Background: Though echocardiography is the cornerstone of cardiac function assessment in specialized practice, there is a lack of point-of-care tools for immediate evaluation of left ventricular ejection fraction (LVEF) in current practice facilitating the early identification of patients at risk for heart failure who may benefit from further echocardiographic evaluation. Aims: To develop and validate artificial intelligence (AI) models to identify reduced LVEF from a single 12-lead ECG using a smartphone application. Methods: We sourced all ECGs and transthoracic echocardiograms (TTEs) recorded between 2011 and 2021. ECGs were paired with TTEs conducted within a 24-hour window and were randomly divided into the model development dataset (50%) and validation dataset (50%). Two AI-ECG models were created: one to detect LVEF ≤40% and another for LVEF <50%, following the AHA definition of heart failure. These models were coupled with smartphone-based ECG digitization technology. Performance metrics included area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score. Results: A total of 1,205,370 ECGs and 291,433 TTEs were collected and paired, resulting in 109,809 ECG-TTE pairs from 56,236 unique patients. The validation dataset consisted of 25,510 distinct TTE-ECG pairs (25,510 patients). Prevalence of LVEF≤40% and LVEF<50% was 5.4% and 7.9% respectively. The LVEF≤40% model demonstrated an AUC of 0.963 (95% CI: 0.959-0.966), sensitivity 0.924 (95% CI: 0.91-0.937), specificity 0.887 (95% CI: 0.883-0.891), and F1-score of 0.474 (95% CI: 0.457-0.490). PPV and NPV were 0.318 (95% CI: 0.304-0.333) and 0.995 (95% CI: 0.994-0.996) respectively. Performance of the LVEF <50% model shows an AUC of 0.952 (95% CI: 0.947-0.956), with a slightly lower sensitivity of 0.899 (95% CI: 0.886-0.912), specificity of 0.875 (95% CI: 0.871-0.879), PPV of 0.382 (95% CI: 0.368-0.395), NPV of 0.99 (95% CI: 0.989-0.992), and F1-score of 0.536 (0.521-0.55). Conclusion: The smartphone-integrated AI model can reliably detect reduced LVEF from standard 12-lead ECGs. Our findings suggest these single 12 lead-ECG based models could serve as a point-of-care screening tool to identify such patients benefiting from further echocardiographic evaluation and consequent management acceleration.

We propose a simple scheme for merging two neural networks trained with different starting initialization into a single one with the same size as the original ones. We do this by carefully selecting channels from each input network. Our procedure might be used as a finalization step after one tries multiple starting seeds to avoid an unlucky one. We also show that training two networks and merging them leads to better performance than training a single network for an extended period of time.

Background: The long-term sequelae of coronavirus disease 2019 (COVID-19) significantly affects quality of life (QoL) in disease survivors. Delayed development of the adaptive immune response is associated with more severe disease and a worse prognosis in COVID-19. The effects of delayed immune response on COVID-19 sequelae and QoL are unknown. Methods: We conducted a prospective study to assess the relationship between the delayed antibody response in the acute phase of infection in naïve unvaccinated patients suffering from severe or critical COVID-19 and their QoL 12 months after hospital discharge. The 12-item Short Form Survey (SF-12) questionnaire was used for assessment of QoL. The SF-12 evaluates both mental and physical components of QoL, incorporating a mental component score (MCS-12) and a physical component score (PCS-12). A delayed antibody response was defined as testing negative for anti-spike SARS-CoV-2 antibodies at the time of hospital admission. Results: The study included 274 patients (154 men and 120 women). Of the enrolled patients, 144 had a delayed immune response. These patients had a significantly lower MCS-12 (p = 0.002), but PCS-12 (p = 0.397) was not significantly different at the 12-month follow-up compared to patients with positive anti-spike SARS-CoV-2 antibodies. The MCS-12 at the time of follow-up was negatively associated with delayed antibody response irrespective of possible confounders (p = 0.006; B = 3.609; ηp2 = 0.035; 95% CI = 1.069–6.150). An MSC-12 below 50 points at the time of follow-up was positively associated with delayed antibody response (p = 0.001; B = 1.092; OR = 2.979; 95% CI = 1.554–5.711). Conclusions: This study confirmed that, in patients with severe and critical COVID-19, a negative result for anti-spike SARS-CoV-2 antibodies at the time of hospital admission is associated with a lower mental component of QoL in unvaccinated patients naïve to COVID-19 one year after hospital discharge.

Pruning large language models (LLMs) is a challenging task due to their enormous size. The primary difficulty is fine-tuning the model after pruning, which is needed to recover the lost performance caused by dropping weights. Recent approaches have either ignored fine-tuning entirely, focusing on efficient pruning criteria, or attempted layer-wise weight updates, preserving the behavior of each layer. However, even layer-wise weight updates can be costly for LLMs, and previous works have resorted to various approximations. In our paper, we propose a fast and effective weight update algorithm for pruned layers based on the Alternating Direction Method of Multipliers (ADMM). We further extend it with a simple gradual pruning mask selection and achieve state-of-the-art pruning performance across a wide range of LLMs.

Base calling in nanopore sequencing is a difficult and computationally intensive problem, typically resulting in high error rates. In many applications of nanopore sequencing, analysis of raw signal is a viable alternative. Dynamic time warping (DTW) is an important building block for raw signal analysis. In this paper, we propose several improvements to DTW class of algorithms to better account for specifics of nanopore signal modeling. We have implemented these improvements in a new signal-to-reference alignment tool Nadavca. We demonstrate that Nadavca alignments improve unsupervised methylation detection over Tombo. We also demonstrate that by providing additional information about the discriminative power of positions in the signal, an otherwise unsupervised method can approach the accuracy of supervised models. Availability and implementation Nadavca is available under MIT license at https://github.com/fmfi-compbio/nadavca . Nanopore sequencing data sets are available from ENA bioproject PRJEB64246. Jaminaea angkorensis reference genome assembly is available from Zenodo https://doi.org/10.5281/zenodo.8145315 .

BACKGROUND:The electrocardiogram (ECG) is one of the most accessible and comprehensive diagnostic tools used to assess cardiac patients at the first point of contact. Despite advances in computerized interpretation of the electrocardiogram (CIE), its accuracy remains inferior to physicians. This study evaluated the diagnostic performance of an artificial intelligence (AI)-powered ECG system and compared its performance to current state-of-the-art CIE. METHODS:An AI-powered system consisting of 6 deep neural networks (DNN) was trained on standard 12‑lead ECGs to detect 20 essential diagnostic patterns (grouped into 6 categories: rhythm, acute coronary syndrome (ACS), conduction abnormalities, ectopy, chamber enlargement and axis). An independent test set of ECGs with diagnostic consensus of two expert cardiologists was used as a reference standard. AI system performance was compared to current state-of-the-art CIE. The key metrics used to compare performances were sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score. RESULTS:A total of 932,711 standard 12‑lead ECGs from 173,949 patients were used for AI system development. The independent test set pooled 11,932 annotated ECG labels. In all 6 diagnostic categories, the DNNs achieved high F1 scores: Rhythm 0.957, ACS 0.925, Conduction abnormalities 0.893, Ectopy 0.966, Chamber enlargement 0.972, and Axis 0.897. The diagnostic performance of DNNs surpassed state-of-the-art CIE for the 13 out of 20 essential diagnostic patterns and was non-inferior for the remaining individual diagnoses. CONCLUSIONS:Our results demonstrate the AI-powered ECG model's ability to accurately identify electrocardiographic abnormalities from the 12‑lead ECG, highlighting its potential as a clinical tool for healthcare professionals.

Introduction: The electrocardiogram (ECG) is a widely available diagnostic tool for evaluating cardiac patients. Although automated ECG interpretation has made significant progress, it has yet to match the accuracy demonstrated by physicians. Hypothesis: In this study, we hypothesized that an artificial intelligence based ECG system can achieve comparable performance to physicians in accurately identifying 20 essential ECG patterns. Methods: An AI-powered system comprising six deep neural networks (DNNs) was trained to identify 20 diagnostic patterns from 12-lead ECGs categorized into six groups: rhythm, infarction, conduction abnormalities, ectopy, chamber enlargement, and axis. An independent test set with the consensus of two expert cardiologists was used as a reference standard. We compared the system's performance to that of three General Practitioners (GPs) and six individual cardiologists, using F1 scores as the evaluation metric. Results: The AI system was trained on 932,711 standard 12-lead ECGs from 173,949 patients. The independent test set comprised 11,932 annotated ECG labels. Figure 1 shows the respective F1 scores of the DNNs, average GP and average cardiologist as follows: Rhythm: 0.957 vs. 0.771 vs. 0.905; Infarction: 0.925 vs. 0.780 vs. 0.852; Conduction abnormalities: 0.893 vs. 0.714 vs. 0.851; Ectopy: 0.966 vs. 0.896 vs. 0.951; Chamber enlargement: 0.972 vs. 0.562 vs. 0.773; Axis: 0.897 vs. 0.601 vs. 0.685. The AI system's diagnostic performance exceeded that of GPs and was on par with cardiologists for all individual diagnostic patterns. Conclusions: The AI-powered ECG system is able to accurately identify electrocardiographic abnormalities from the 12-lead ECG, highlighting its potential as a clinical tool for healthcare professionals.

ABSTRACT Background One third of Non-ST-elevation myocardial infarction (NSTEMI) patients present with an acutely occluded culprit coronary artery (occlusion myocardial infarction [OMI]), which is associated with poor short and long-term outcomes due to delayed identification and consequent delayed invasive management. We sought to develop and validate a versatile artificial intelligence (AI)-model detecting OMI on single standard 12-lead electrocardiograms (ECGs) and compare its performance to existing state-of-the-art diagnostic criteria. Methods An AI model was developed using 18,616 ECGs from 10,692 unique contacts (22.9% OMI) of 10,543 patients (age 66±14 years, 65.9% males) with acute coronary syndrome (ACS) originating from an international online database and a tertiary care center. This AI model was tested on an international test set of 3,254 ECGs from 2,263 unique contacts (20% OMI) of 2,222 patients (age 62±14 years, 67% males) and compared with STEMI criteria and annotations of ECG experts in detecting OMI on 12-lead ECGs using sensitivity, specificity, predictive values and time to OMI diagnosis. OMI was based on a combination of angiographic and biomarker outcomes. Results The AI model achieved an area under the curve (AUC) of 0.941 (95% CI: 0.926-0.954) in identifying the primary outcome of OMI, with superior performance (accuracy 90.7% [95% CI: 89.5-91.9], sensitivity 82.6% [95% CI: 78.9-86.1], specificity 92.8 [95% CI: 91.5-93.9]) compared to STEMI criteria (accuracy 84.9% [95% CI: 83.5-86.3], sensitivity 34.4% [95% CI: 30.0-38.8], specificity 97.6% [95% CI: 96.8-98.2]) and similar performance compared to ECG experts (accuracy 91.2% [95% CI: 90.0-92.4], sensitivity 75.9% [95% CI: 71.9-80.0], specificity 95.0 [95% CI: 94.0-96.0]). The average time from presentation to a correct diagnosis of OMI was significantly shorter when relying on the AI model compared to STEMI criteria (2.0 vs. 4.9 hours, p<0.001). Conclusions The present novel ECG AI model demonstrates superior accuracy and earlier diagnosis of AI to detect acute OMI when compared to the STEMI criteria. Its external and international validation suggests its potential to improve ACS patient triage with timely referral for immediate revascularization. CLINICAL PERSPECTIVE What is new? A novel artificial intelligence (AI) model detecting acute occluded coronary artery (OMI) using standard 12-lead electrocardiograms (ECGs) was developed from an international cohort. The OMI AI model is the first of its kind to be validated in an external international cohort of patients using an objective angiographically confirmed endpoint of OMI. Our study demonstrated the OMI AI models superior accuracy in identifying OMI and shorter time to correct diagnosis compared to standard of care STEMI criteria. What are the clinical implications? The OMI AI model has the potential to improve ACS triage and clinical decision-making by enabling timely and accurate detection of OMI in NSTEMI patients. The robustness and versatility of the OMI AI model indicate its potential for real-world clinical implementation in ECG devices from multiple vendors. Prospective studies are essential to evaluate the efficacy of the OMI AI model and its impact on patient outcomes in real-world settings.

Introduction: One third of Non-ST-elevation myocardial infarction (NSTEMI) patients present with an acutely occluded culprit coronary artery (OMI), associated with poor short and long-term outcomes due to delayed identification and consequent delayed invasive management. Hypothesis: We sought to develop and validate an artificial intelligence (AI) model detecting OMI on single standard 12-lead electrocardiograms (ECGs) and compare its performance to existing standard criteria. Methods: An AI model was developed using 18,616 ECGs from 10,692 unique contacts (22.9% OMI) of 10,543 patients (age 66±14 years, 65.9% males) with acute coronary syndrome (ACS) from an international cohort. This AI model was tested on an international test set of 3,254 ECGs from 2,263 unique contacts (20% OMI) of 2,222 patients (age 62±14 years, 67% males) and compared with STEMI criteria and annotations of ECG experts in detecting OMI on 12-lead ECGs using sensitivity, specificity, predictive values and time to OMI diagnosis. OMI was based on a combination of angiographic and biomarker outcomes. Results: The AI model achieved an area under the curve (AUC) of 0.941 (95% CI: 0.926-0.954) in identifying the primary outcome of OMI, with superior performance (accuracy 90.7% [95% CI: 89.5-91.9], sensitivity 82.6% [95% CI: 78.9-86.1], specificity 92.8 [95% CI: 91.5-93.9]) compared to STEMI criteria (accuracy 84.9% [95% CI: 83.5-86.3], sensitivity 34.4% [95% CI: 30.0-38.8], specificity 97.6% [95% CI: 96.8-98.2]) and similar performance compared to ECG experts (accuracy 91.2% [95% CI: 90.0-92.4], sensitivity 75.9% [95% CI: 71.9-80.0], specificity 95.0 [95% CI: 94.0-96.0]) ( Figure 1 ). Conclusions: The present novel ECG AI model demonstrates superior accuracy in detecting acute OMI when compared to the STEMI criteria. Its external and international validation suggests its potential to improve ACS patient triage with timely referral for immediate revascularization.

To connect digital and physical devices a highly connected global network is termed as Internet. The objective of internet is extended by Internet of Thing (IoT) by connecting devices from different domain. Millions of devices are used for information collection and communication in IoT. Integration of nanotechnology with IoT termed as Internet of Nano Things (IoNT) and plays an Important role in growth of compact devices. The requirement of feature and functionality in compact, handheld and battery operated devices increasing day by day. The battery life is an important factor for these devices. The battery life of these devices can be increased by reducing the power consumption. Single electron devices can play an important role in limiting the power consumption of these devices. In this paper a Si dot Single Electron Transistor is designed with nano scale channel length. Si dot is placed between gate and island to improve electrical properties of device. The characteristics of Si dot SET is compared with a SET without Si dot and capacitance is calculated for both the devices. It is found that device with Si dot possess higher charging energy so it is capable of room temperature operation. This paper also discusses the role of nano devices in IoNT as well as application areas and challenges of IoNT.

In nanopore sequencing, electrical signal is measured as DNA molecules pass through the sequencing pores. Translating these signals into DNA bases (base calling) is a highly non-trivial task, and its quality has a large impact on the sequencing accuracy. The most successful nanopore base callers to date use convolutional neural networks (CNN) to accomplish the task. Convolutional layers in CNNs are typically composed of filters with constant window size, performing best in analysis of signals with uniform speed. However, the speed of nanopore sequencing varies greatly both within reads and between sequencing runs. Here, we present dynamic pooling, a novel neural network component, which addresses this problem by adaptively adjusting the pooling ratio. To demonstrate the usefulness of dynamic pooling, we developed two base callers: Heron and Osprey. Heron improves the accuracy beyond the experimental high-accuracy base caller Bonito developed by Oxford Nanopore. Osprey is a fast base caller that can compete in accuracy with Guppy high-accuracy mode, but does not require GPU acceleration and achieves a near real-time speed on common desktop CPUs. Availability: https://github.com/fmfi-compbio/osprey , https://github.com/fmfi-compbio/heron .

The association between COVID-19 severity and antibody response has not been clearly determined. We aimed to assess the effects of antibody response to SARS-CoV-2 S protein at the time of hospital admission on in-hospital and longitudinal survival. Methods: A prospective observational study in naive hospitalised COVID-19 patients. The presence of anti-S SARS-CoV-2 IgM and IgG was evaluated using a lateral flow assay at the time of admission. The patients were followed up for 8–30 months to assess survival. We recruited 554 patients (330 men and 224 women). Overall, 63.0% of the patients had positive IgG or IgM anti-S SARS-CoV-2 antibodies at the time of hospital admission. In the univariate analysis, the patients with negative anti-S SARS-CoV-2 IgM and IgG antibodies were referred to the hospital sooner, had lower CRP and D-dimer concentrations, and were hospitalised longer. They were also more likely to be admitted to an intensive care unit and more often received baricitinib treatment. During their hospital stay, 8.5% of the antibody-positive and 22.3% of the antibody-negative patients died (p = 0.0001). The median duration of the follow-up was 21 months. During the follow-up after hospital discharge, 3.6% of antibody-positive and 9.1% of antibody-negative patients died (p = 0.027). In the multivariate analysis, the negative anti-S SARS-CoV-2 antibodies were associated with a higher risk of in-hospital death (OR 3.800; 95% CI 1.844–7.829; p = 0.0001) and with a higher risk of death during follow-up (OR 2.863; 95% CI 1.110–7.386; p = 0.030). These associations were independent of age, the time from symptom onset to hospital admission, CRP, D-Dimer, the number of comorbidities, disease severity at the time of hospital admission, and baricitinib therapy. Our study concludes that negative anti-S SARS-CoV-2 IgM and IgG at the time of admission are associated with higher in-hospital mortality and cause a higher risk of all-cause death during follow-up after discharge.

In this study, we ascertain the associations between BCG vaccination policies and progression of COVID-19 through analysis of various time-adjusted indicators either directly extracted from the incidence and death reports, or estimated as parameters of disease progression models. We observe weak correlation between BCG vaccination status and indicators related to disease reproduction characteristics. We did not find any associations with case fatality rates (CFR), but the differences in CFR estimates are at present likely dominated by differences in testing and case reporting between countries.

We developed a new base caller DeepNano-coral for nanopore sequencing, which is optimized to run on the Coral Edge Tensor Processing Unit, a small USB-attached hardware accelerator. To achieve this goal, we have designed new versions of two key components used in convolutional neural networks for speech recognition and base calling. In our components, we propose a new way of factorization of a full convolution into smaller operations, which decreases memory access operations, memory access being a bottleneck on this device. DeepNano-coral achieves real-time base calling during sequencing with the accuracy slightly better than the fast mode of the Guppy base caller and is extremely energy efficient, using only 10W of power. Availability: https://github.com/fmfi-compbio/coral-basecaller