100 research outputs found
Fortschritte für die Herzgesundheit: Vertrauenswürdige und praxisnahe Deep-Learning-Ansätze zur Analyse von 12-Kanal-EKGs
No full textThe electrocardiogram (ECG) is a common diagnostic tool that is crucial for the assessment of cardiac status and is usually analyzed manually with basic rule-based algorithm support. In recent years, however, a growing number of studies have highlighted the enormous diagnostic potential of combining ECG with machine learning and deep learning techniques. Despite remarkable progress, their application still faces significant challenges and limitations, including the lack of transparency of the inner workings of these algorithms, insufficient robustness against physiological noise, or the scarcity of (labeled) data in the field of cardiology. In this work, we present approaches and methods to address these challenges. We test machine learning models, both those that operate on raw ECGs and those that take pre-extracted features as input, by analyzing their decision-making process and then comparing it to cardiologists’ diagnosis rules to ensure correct model behavior. By extracting and modeling physiological noise, we evaluate and improve the robustness of deep learning models to realistic noise. In addition, we propose methods to utilize unlabeled data via self-supervised learning paradigms for enhanced performance and improved label efficiency. Finally, we investigate state-space deep learning models that are well suited for time series data and show that incorporating patient metadata into the decision-making process leads to an astonishing increase in accuracy in the diagnosis of heart disease. Together, these approaches promote a unified vision of deep learning for 12-lead ECG classification, representing an important first step toward the real-world application of deep learning for cardiology in clinical settings.Das Elektrokardiogramm (EKG) ist ein gängiges Diagnoseinstrument, das für die Beurteilung des Herzzustands von entscheidender Bedeutung ist und in der Regel manuell oder mit Hilfe einfacher regelbasierter Algorithmen analysiert wird. In den letzten Jahren hat jedoch eine wachsende Zahl von Studien das enorme diagnostische Potenzial der Kombination von EKG mit maschinellem Lernen und Deep-Learning-Techniken aufgezeigt. Trotz bemerkenswerter Fortschritte stößt ihre Anwendung noch immer auf erhebliche Herausforderungen und Einschränkungen, darunter die mangelnde Transparenz der inneren Funktionsweise dieser Algorithmen, die unzureichende Robustheit gegenüber physiologischem Rauschen oder die Knappheit an (markierten) Daten im Bereich der Kardiologie. In dieser Arbeit stellen wir Ansätze und Methoden vor, um diesen Herausforderungen zu begegnen. Wir testen maschinelle Lernmodelle, sowohl solche, die mit Roh-EKGs arbeiten, als auch solche, die zuvor extrahierte Merkmale als Eingabe verwenden, indem wir ihren Entscheidungsprozess analysieren und ihn dann mit den Diagnoseregeln von Kardiologen vergleichen, um ein korrektes Modellverhalten sicherzustellen. Durch die Extraktion und Modellierung von physiologischem Rauschen können wir die Robustheit von Deep-Learning-Modellen gegenüber realistischem Rauschen bewerten und verbessern. Darüber hinaus schlagen wir Methoden vor, um nicht annotierte Daten über selbstüberwachte Lernparadigmen zu nutzen, um die Leistung und die Dateneffizienz zu verbessern. Schließlich untersuchen wir State-Space-Deep-Learning-Modelle, die sich gut für Zeitreihendaten eignen, und zeigen, dass die Einbeziehung von Patienten-Metadaten in den Entscheidungsprozess zu einer erstaunlichen Steigerung der Genauigkeit bei der Diagnose von Herzerkrankungen führt. Zusammen fördern diese Ansätze eine einheitliche Vision von Deep Learning für die 12-Kanal-EKG-Klassifizierung und stellen einen wichtigen ersten Schritt in Richtung der realen Anwendung von Deep Learning für die Kardiologie im klinischen Umfeld dar
PTB-XL (v1.0.1) Soft Segmentations (Delineation)
No full textOverview
The dataset provides (soft) segmentation/delineation masks for the PTB-XL ECG dataset. The underlying original dataset (v1.0.1) is available from PhysioNet (doi.org/10.13026/kfzx-aw45) For further details on the orginal dataset, consult the dataset descriptor (Wagner, P., Strodthoff, N., Bousseljot, RD. et al. PTB-XL, a large publicly available electrocardiography dataset. Sci Data 7, 154 (2020). https://doi.org/10.1038/s41597-020-0495-6).
Dataset
The dataset was obtained by training a U-net segmentation model using
ECGDeli (Pilia, N., Nagel, C., et al. ECGdeli - An open source ECG delineation toolbox for MATLAB. SoftwareX 13, 100639 (2021).
https://doi.org/10.1016/j.softx.2020.100639.).
as initial labels increased by adding intermediate segments. Details on the model architecture will be described in a forthcoming article (Wagner, P., Mehari, T. et al.).
Dataset format
In the folder ptbxl_segmentations_8bit, we provide files in numpy format (8 bit unsigned integer) with filenames corresponding to the ECG ID of the corresponding sample in PTB-XL. Each file contains an array of shape 1000x12x24, where the first axis corresponds to time steps in the sequence (10s at 100 Hz), the second refers to the 12 ECG leads (['I', 'II', 'III', 'aVR', 'aVL', 'aVF', 'V1', 'V2', 'V3', 'V4', 'V5', 'V6']) and the third axis refers to ['P-Onset', '(p', 'P-Peak', 'p)', 'P-Offset', 'pq', 'QRS-Onset', '(q', 'Q-Peak', 'qr', 'R-Peak', 'rs', 'S-Peak', 'q)', 'QRS-Offset', 'qj', 'J-point', 'jt', 'T-Onset', '(t', 'T-Peak', 't)', 'T-Offset', 'tp']. For every time step and channel it provides a soft output distribution over the 24 output classes of the segmentation model
Towards quantitative precision for ECG analysis: Leveraging state space models, self-supervision and patient metadata
No full text
Frequency Hopping in LTE Uplink
No full textIn the 3GPP LTE, different radio resource management (RRM) techniques have been proposed in order to improve the uplink performance. Frequency hopping is one of the techniques that can be used to improve the uplink performance by providing frequency diversity and interference averaging. The hopping can be between subframes (inter-subframe) or within a subframe (intra-subframe). 3GPP specifies two types of frequency hopping for the LTE uplink, hopping based on explicit hopping information in the scheduling grant and sub-band based hopping according to cell-specific hopping and mirroring patterns. In this master’s thesis, theoretical discussion on the frequency hopping schemes is carried out followed by dynamic simulations in order to evaluate the performance gain of frequency hopping. Based on the theoretical analysis, the second type of hopping is selected for detailed study. As a baseline for comparison, dynamic frequency domain scheduling with random frequency resource allocation has been used. Single cell and multi-cell scenarios have been simulated with VoIP traffic model using user satisfaction as a performance metric. The simulation results show that frequency hopping improves the uplink performance by providing frequency diversity in the single cell scenario and both frequency diversity and interference averaging in the multi-cell scenario. The gains in using the hopping schemes were reflected as VoIP capacity (the number of satisfied users) improvement. In this study, the performance of the selected hopping schemes under different hopping parameters is also [email protected] [email protected]
Self-supervised representation learning from 12-lead ECG data
No full textArt. 105114, 8 S.Clinical 12-lead electrocardiography (ECG) is one of the most widely encountered kinds of biosignals. Despite the increased availability of public ECG datasets, label scarcity remains a central challenge in the field. Self-supervised learning represents a promising way to alleviate this issue. This would allow to train more powerful models given the same amount of labeled data and to incorporate or improve predictions about rare diseases, for which training datasets are inherently limited. In this work, we put forward the first comprehensive assessment of self-supervised representation learning from clinical 12-lead ECG data. To this end, we adapt state-of-the-art self-supervised methods based on instance discrimination and latent forecasting to the ECG domain. In a first step, we learn contrastive representations and evaluate their quality based on linear evaluation performance on a recently established, comprehensive, clinical ECG classification task. In a second step, we analyze the impact of self-supervised pretraining on finetuned ECG classifiers as compared to purely supervised performance. For the best-performing method, an adaptation of contrastive predictive coding, we find a linear evaluation performance only 0.5% below supervised performance. For the finetuned models, we find improvements in downstream performance of roughly 1% compared to supervised performance, label efficiency, as well as robustness against physiological noise. This work clearly establishes the feasibility of extracting discriminative representations from ECG data via self-supervised learning and the numerous advantages when finetuning such representations on downstream tasks as compared to purely supervised training. As first comprehensive assessment of its kind in the ECG domain carried out exclusively on publicly available datasets, we hope to establish a first step towards reproducible progress in the rapidly evolving field of representation learning for biosignals.14
Conservation tillage systems and water productivity implications for smallholder farmers in semi-arid Ethiopia
No full textConservation tillage systems have been adopted by farmers in many countries to solve the problem of land degradation and declining water productivity. However, direct application of such tillage systems was not possible among resource poor smallholder farmers in semi arid areas of Ethiopia. Problems such as shortage of rainfall, cost of herbicides, cost of implements and the small seeded crop, tef, which can not be planted in rows required development of locally adapted conservation tillage systems. This book presents the problems of traditional tillage systems and the results of tests carried out on appropriate conservation tillage implements and systems for smallholder farmers in semi arid regions of Ethiopia. The traditional tillage implement, the Maresha Plow and the related tillage systems were identified to be the main causes of repeated and cross plowing that led to land degradation and reduced water productivity. The Maresha modified implements were found to be suitable to undertake conservation tillage systems while being simple, light and affordable. Two types of tillage systems developed for maize and tef were found to reduce surface runoff, increase availability of water to crops and increase yields. The way forward and recommended areas of future research are also presented. More information: http://www.taylorandfrancis.co.ukCivil Engineering and Geoscience
GGE Bi Plot Analysis of Genotype X Environment Interaction and Grain Yield Stability of Bread Wheat (Triticum aestivum L.) Genotypes in Ethiopia
No full textThere is limitation of information on G x EI of bread wheat genotypes in Ethiopia. The study cried out with objectives to estimate Genotype x Environment interaction and stability of bread wheat genotypes in Ethiopia. Thirty Bread wheat genotypes were evaluated by Alpha lattice design using three replications at eight locations in Ethiopia. The mean grain yield of genotypes across environments was 4.53 ton ha-1. Bread wheat grain yield was significantly affected by the E, G and G x E interaction. Environment, G x E interaction and genotype explained 45.59%, 25.37% and 2.59% of the total (G + E + GEI) variation respectively. Genotype ETBW71942 (3), ETBW7038 (9), ETBW8511 (1) and ETBW8512 (14) were considered specifically adapted. Considering simultaneously yield and stability, Genotype ETBW7871 (15), ETBW7058 (11), ETBW8513 (16) and ETBW7101 (25) showed the best performances. Keywords: genotype; environment; genotype x environment interaction; Stability
Dithered backprop: A sparse and quantized backpropagation algorithm for more efficient deep neural network training
No full textComparative Evaluation of Model Sensitivity, Calibration, and Parameter Uncertainty in Streamflow Simulation Using SWAT and HBV Light in the Geba Watershed, Ethiopia
No full textEvaluating and simulating streamflow is extremely useful for managing water resources in almost all regions, particularly in arid ones. This study focuses on the performance of two hydrological models, HBV Light and SWAT, in streamflow simulation in northern Ethiopia. The models were evaluated using an ensemble modeling approach, which integrated Monte Carlo simulations for HBV Light and SWAT, while SWAT-CUP was used for calibration and validation. During the sensitivity analysis, key parameters controlling the model outputs were identified. For HBV Light, the parameters, K2, MAXBAS, and BETA, reflect subsurface processes, whereas SWAT, CN2, GWQMN, and SOL_AWC were used to control surface runoff. During calibration and validation, SWAT demonstrated statistically superior performance in modeling streamflow (R²=0.73, NSE=0.81) and (R2 =0.72, NSE=0.72) respectively. While HBV Light recorded a performance of (R²=0.71, NSE=0.70) during calibration and (R2= 0.71, NSE= 0.71) during validation,which was closer to the observed streamflow. This indicates during the validation phase, SWAT still performed better but HBV Light demonstrated narrower predictive uncertainties at 95% along with more identifiability of the parameters that reduced the problem of equifinality. The bottom line of this case was that SWAT was statistically better, while HBV Light was more transparent and reliable with uncertainties. All things considered, both models could simulate streamflow, but their differences suggested that context-based choice would be optimal. While predictive consistency and uncertainty portrayal were most important, HBV Light was advantageous, and SWAT was better suited for use in cases that rank calibration accuracy above all else. The hydrology of the catchment could be better understood if streamline decision-making for the effective and sustainable management of water resources in the Geba Catchment and similar semi-arid areas were combined, or if multi-model ensembles were utilized.
Keywords: Streamflow simulation, Hydrological modeling, HBV Light model, SWAT model, Uncertainty analysis, Parameter sensitivit
- …
