Variational Autoencoders for Data Augmentation in Clinical Studies

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(15), P. 8793 - 8793

Published: July 30, 2023

Sample size estimation is critical in clinical trials. A sample of adequate can provide insights into a given population, but the collection substantial amounts data costly and time-intensive. The aim this study was to introduce novel augmentation approach field trials by employing variational autoencoders (VAEs). Several forms VAEs were developed used for generation virtual subjects. Various types explored employed production individuals, several different scenarios investigated. VAE-generated exhibited similar performance original data, even cases where small proportion them (e.g., 30–40%) reconstruction generated data. Additionally, showed higher statistical power than high variability. This represents an additional advantage use situations variability, as they act noise reduction. application be useful tool decreasing required and, consequently, reducing costs time involved. Furthermore, it aligns with ethical concerns surrounding human participation

Language: Английский

---

An intelligent deep feature based metabolism syndrome prediction system for sleep disorder diseases

Multimedia Tools and Applications, Journal Year: 2023, Volume and Issue: 83(17), P. 51267 - 51290

Published: Nov. 10, 2023

Language: Английский

---

Missing Data Statistics Provide Causal Insights into Data Loss in Diabetes Health Monitoring by Wearable Sensors

Sensors, Journal Year: 2024, Volume and Issue: 24(5), P. 1526 - 1526

Published: Feb. 27, 2024

Background: Data loss in wearable sensors is an inevitable problem that leads to misrepresentation during diabetes health monitoring. We systematically investigated missing data get causal insight into the mechanisms leading data. Methods: Two-week-long from a continuous glucose monitor and Fitbit activity tracker recording heart rate (HR) step count free-living patients with type 2 mellitus were used. The gap size distribution was fitted Planck test for not at random (MNAR) difference between distributions tested Chi-squared test. Significant dispersion over time Kruskal–Wallis Dunn post hoc analysis. Results: 77 subjects resulted 73 cleaned glucose, 70 HR 68 recordings. sizes followed distribution. frequency differed significantly (p < 0.001), therefore MNAR. In more found night (23:00–01:00), count, measurement days 6 7 0.001). both cases, caused by insufficient of synchronization. Conclusions: Our novel approach investigating statistics revealed CGM

Language: Английский

---

Wireless Mouth Motion Recognition System Based on EEG-EMG Sensors for Severe Speech Impairments

Sensors, Journal Year: 2024, Volume and Issue: 24(13), P. 4125 - 4125

Published: June 25, 2024

This study aims to demonstrate the feasibility of using a new wireless electroencephalography (EEG)–electromyography (EMG) wearable approach generate characteristic EEG-EMG mixed patterns with mouth movements in order detect distinct movement for severe speech impairments. paper describes method detecting based on signal processing technology suitable sensor integration and machine learning applications. examines relationship between motion brainwave an effort develop nonverbal interfacing people who have lost ability communicate, such as paralysis. A set experiments were conducted assess efficacy proposed feature selection. It was determined that classification meaningful. signals also collected during silent mouthing phonemes. few-shot neural network trained classify phonemes from signals, yielding accuracy 95%. technique data collection bioelectrical phoneme recognition proves promising avenue future communication aids.

Language: Английский

---

An Improved Approach for Atrial Fibrillation Detection in Long-Term ECG Using Decomposition Transforms and Least-Squares Support Vector Machine

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(22), P. 12187 - 12187

Published: Nov. 9, 2023

Atrial fibrillation is a common heart rhythm disorder that now becoming significant healthcare challenge as it affects more and people in developed countries. This paper proposes novel approach for detecting this disease. For purpose, we examined the ECG signal by QRS complexes then selecting 30 successive R-peaks analyzing atrial activity segment with variety of indices, including entropy change, variance wavelet transform distribution energy bands determined dual-Q tunable Q-factor coefficients Hilbert ensemble empirical mode decomposition. These transformations provided vector 21 features characterized relevant part electrocardiography signal. The MIT-BIH Fibrillation Database was used to evaluate proposed method. Then, using K-fold cross-validation method, sets were fed into LS-SVM SVM classifiers trilayered neural network classifier. Training test subsets set up avoid sampling from single participant maintain balance between classes. In addition, individual classification quality scores analyzed each determine dependencies on subject. results obtained during testing procedure showed sensitivity 98.86%, positive predictive value 99.04%, accuracy 98.95%.

Language: Английский

---

Heart Sound Classification Using Harmonic and Percussive Spectral Features from Phonocardiograms with a Deep ANN Approach

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(22), P. 10201 - 10201

Published: Nov. 6, 2024

Cardiovascular diseases (CVDs) are a leading cause of mortality worldwide, with particularly high burden in India. Non-invasive methods like Phonocardiogram (PCG) analysis capture the acoustic activity heart. This holds significant potential for early detection and diagnosis heart conditions. However, complexity variability PCG signals pose considerable challenges accurate classification. Traditional signal analysis, including time-domain, frequency-domain, time-frequency domain techniques, often fall short capturing intricate details necessary reliable diagnosis. study introduces an innovative approach that leverages harmonic–percussive source separation (HPSS) to extract distinct harmonic percussive spectral features from signals. These then utilized train deep feed-forward artificial neural network (ANN), classifying conditions as normal or abnormal. The methodology involves advanced digital processing techniques applied recordings PhysioNet 2016 dataset. feature set comprises 164 attributes, Chroma STFT, CENS, Mel-frequency cepstral coefficients (MFCCs), statistical features. refined using ROC-AUC selection method ensure optimal performance. ANN model was rigorously trained validated on balanced Techniques such noise reduction outlier were used improve training. proposed achieved validation accuracy 93.40% sensitivity specificity rates 82.40% 80.60%, respectively. results underscore effectiveness harmonic-based robustness sound research highlights deploying models non-invasive cardiac diagnostics, resource-constrained settings. It also lays groundwork future advancements analysis.

Language: Английский

---

A Novel Diagnostic Framework for Breast Cancer: Combining Deep Learning with Mammogram-DBT Feature Fusion

Results in Engineering, Journal Year: 2024, Volume and Issue: unknown, P. 103836 - 103836

Published: Dec. 1, 2024

Language: Английский

---

Interpretable Single-dimension Outlier Detection (ISOD): An Unsupervised Outlier Detection Method Based on Quantiles and Skewness Coefficients

Applied Sciences, Journal Year: 2023, Volume and Issue: 14(1), P. 136 - 136

Published: Dec. 22, 2023

A crucial area of study in data mining is outlier detection, particularly the areas network security, credit card fraud industrial flaw etc. Existing detection algorithms, which can be divided into supervised methods, semi-supervised and unsupervised suffer from missing labeled data, curse dimensionality, low interpretability, To address these issues, this paper, we present an method based on quantiles skewness coefficients called ISOD (Interpretable Single dimension Outlier Detection). first fulfils empirical cumulative distribution function before computing quantile each dimension. Finally, it outputs score. This paper’s contributions are as follows: (1) propose algorithm ISOD, has high interpretability scalability; (2) massive experiments benchmark datasets demonstrated superior performance compared with state-of-the-art baselines terms ROC AP.

Language: Английский

---

Generative AI-enabled Knowledge Base Fine-tuning: Enhancing Feature Engineering for Customer Churn

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 4, 2024

Abstract Customers are the most critical component in a business’s success regardless of industry or product. Companies make significant efforts to acquire and, more importantly, retain their existing customers. Customer churn is challenge for businesses, leading financial losses. To address this challenge, understanding customer’s cognitive status, behaviors, and early signs crucial. However, predictive ML-based analysis, being fed with proper features that indicative status behavior, extremely helpful addressing challenge. Having practical analysis relies on well-developed feature engineering process. Previous analytical studies mainly applied approaches leveraged demographic, product usage, revenue alone, there lack research leveraging information-rich content from interactions between customers companies. Considering effectiveness applying domain knowledge human expertise engineering, motivated by our previous work, we propose Churn-related Knowledge Base (ChurnKB) enhance In ChurnKB, leverage textual data mining techniques extracting churn-related texts created customers, e.g., emails chat logs company agents, reviews company’s website, feedback social media. We use Generative AI (GAI) enrich structure ChurnKB regarding related customer feelings, behaviors. also loops crowdsourcing approve validity proposed apply it develop classifier problems.

Language: Английский

---

Disentangled Conditional Variational Autoencoder for Unsupervised Anomaly Detection

2021 IEEE International Conference on Big Data (Big Data), Journal Year: 2024, Volume and Issue: unknown, P. 138 - 143

Published: Dec. 15, 2024

Language: Английский

---