Multistage transfer learning for medical images

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 57(9)

Published: Aug. 6, 2024

Abstract Deep learning is revolutionizing various domains and significantly impacting medical image analysis. Despite notable progress, numerous challenges remain, necessitating the refinement of deep algorithms for optimal performance in This paper explores growing demand precise robust analysis by focusing on an advanced technique, multistage transfer learning. Over past decade, has emerged as a pivotal strategy, particularly overcoming associated with limited data model generalization. However, absence well-compiled literature capturing this development remains gap field. exhaustive investigation endeavors to address providing foundational understanding how approaches confront unique posed insufficient datasets. The offers detailed types, architectures, methodologies, strategies deployed Additionally, it delves into intrinsic within framework, comprehensive overview current state while outlining potential directions advancing methodologies future research. underscores transformative analysis, valuable guidance researchers healthcare professionals.

Language: Английский

---

Video-based beat-by-beat blood pressure monitoring via transfer deep-learning

Applied Intelligence, Journal Year: 2024, Volume and Issue: 54(6), P. 4564 - 4584

Published: March 1, 2024

Abstract Currently, learning physiological vital signs such as blood pressure (BP), hemoglobin levels, and oxygen saturation, from Photoplethysmography (PPG) signal, is receiving more attention. Despite successive progress that has been made so far, continuously revealing new aspects characterizes field a rich research topic. It includes diverse number of critical points represented in signal denoising, data cleaning, employed features, feature format, selection, domain, model structure, problem formulation (regression or classification), combinations. worth noting extensive efforts are devoted to utilizing different variants machine deep models while transfer not fully explored yet. So, this paper, we introducing per-beat rPPG-to-BP mapping scheme based on learning. An interesting representation 1-D PPG 2-D image proposed for enabling powerful off-the-shelf image-based through resolves limitations about training size due strict cleaning. Also, it enhances generalization by exploiting underlying excellent extraction. Moreover, non-uniform distribution (data skewness) partially resolved logarithmic transformation. Furthermore, double cleaning applied contact testing rPPG beats well. The quality the segmented tested checking some related metrics. Hence, prediction reliability enhanced excluding deformed beats. Varying relaxed selecting during intervals highest strength. Based experimental results, system outperforms state-of-the-art systems sense mean absolute error (MAE) standard deviation (STD). STD test decreased 5.4782 3.8539 SBP DBP, respectively. MAE 2.3453 1.6854 results BP estimation real video reveal reaches 8.027882 6.013052 estimated videos 7.052803 5.616028 Graphical abstract Proposed camera-based monitoring

Language: Английский

---

Development of a Novel Light-Sensitive PPG Model Using PPG Scalograms and PPG-NET Learning for Non-Invasive Hypertension Monitoring

Heliyon, Journal Year: 2024, Volume and Issue: 10(21), P. e39745 - e39745

Published: Oct. 23, 2024

Photoplethysmography (PPG) signals provide a non-invasive method for monitoring cardiovascular health, including blood pressure levels, which are critical the early detection and management of hypertension. This study leverages wavelet transformation special purpose deep learning model, enhanced by signal processing normalization, to classify stages from PPG signals. The primary objective is advance hypertension monitoring, improving accuracy efficiency these assessments.

Language: Английский

---

Blood Pressure Estimation with Phonocardiogram on CNN-Based Approach

Computers, materials & continua/Computers, materials & continua (Print), Journal Year: 2024, Volume and Issue: 79(2), P. 1775 - 1794

Published: Jan. 1, 2024

Monitoring blood pressure is a critical aspect of safeguarding an individual's health, as early detection abnormal levels facilitates timely medical intervention, ultimately leading to reduction in mortality rates associated with cardiovascular diseases.Consequently, the development robust and continuous monitoring system holds paramount significance.In context this research paper, we introduce innovative deep learning regression model that harnesses phonocardiogram (PCG) data achieve precise estimation.Our novel approach incorporates convolutional neural network (CNN)-based model, which not only enhances its adaptability spatial variations but also empowers it capture intricate patterns within PCG signals.These advancements contribute significantly overall accuracy estimation.To substantiate effectiveness our proposed method, meticulously gathered signal from 78 volunteers, adhering ethical guidelines Suranaree University Technology (Human Research Ethics number EC-65-78).Subsequently, rigorously preprocessed dataset ensure integrity.We further employed K-fold cross-validation procedure for division alignment, combining resulting datasets CNN estimation.The experimental results are highly promising, yielding Mean Absolute Error (MAE) standard deviation (STD) approximately 10.69 ± 7.23 mmHg systolic 6.89 5.22 diastolic pressure.Our study underscores potential estimation, particularly using signals, paving way practical, non-invasive method broad applicability healthcare domain.Early can facilitate interventions, reducing disease-related rates.

Language: Английский

---

Photoplethysmography Features Correlated with Blood Pressure Changes

Diagnostics, Journal Year: 2024, Volume and Issue: 14(20), P. 2309 - 2309

Published: Oct. 17, 2024

Blood pressure measurement is a key indicator of vascular health and routine part medical examinations. Given the ability photoplethysmography (PPG) signals to provide insights into microvascular bed their compatibility with wearable devices, significant research has focused on using PPG for blood estimation. This study aimed identify specific clinical features that vary different levels. Through literature review 297 publications, we selected 16 relevant studies identified time-dependent associated prediction. Our analysis highlighted second derivative signals, particularly b/a d/a ratios, as most frequently reported predictors systolic pressure. Additionally, from velocity acceleration photoplethysmograms were also notable. In total, 29 analyzed, revealing novel temporal domain show promise further application in

Language: Английский

---

Cuffless Blood Pressure Estimation with Confidence Intervals using Hybrid Feature Selection and Decision Based on Gaussian Process

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(2), P. 1221 - 1221

Published: Jan. 16, 2023

Cuffless blood pressure (BP) monitoring is crucial for patients with cardiovascular disease and hypertension. However, conventional BP monitors provide only single-point estimates without confidence intervals. Therefore, the statistical variability in indistinguishable from intrinsic caused by physiological processes. This study introduced a novel method improving reliability of intervals (CIs) estimations using hybrid feature selection decision based on Gaussian process. F-test robust neighbor component analysis were applied as methods obtaining set highly weighted features to estimate accurate CIs. Akaike’s information criterion algorithm was used select best subset. The performance proposed confirmed through experiments. Comparisons algorithms indicated that provided most CIs estimates. To authors’ knowledge, currently one provides reliable may be concurrently estimating

Language: Английский

---

Temporal complexity in photoplethysmography and its influence on blood pressure

Frontiers in Physiology, Journal Year: 2023, Volume and Issue: 14

Published: Aug. 31, 2023

Objective: The temporal complexity of photoplethysmography (PPG) provides valuable information about blood pressure (BP). In this study, we aim to interpret the stochastic PPG patterns with a model-based simulation, which may help optimize BP estimation algorithms. Methods: classic four-element Windkessel model is adapted in study incorporate BP-dependent compliance profiles. Simulations are performed generate responses pulse and continuous stimuli at various timescales, aiming mimic sudden or gradual hemodynamic changes observed real-life scenarios. To quantify PPG, utilize Higuchi fractal dimension (HFD) autocorrelation function (ACF). These measures provide insights into intricate exhibited by PPG. validate simulation results, recordings BP, stroke volume from 40 healthy subjects were used. Results: Pulse simulations showed that central vascular variation during cardiac cycle, peripheral resistance, output (CO) collectively contributed time delay, amplitude overshoot, phase shift responses. Continuous could be generated random stimuli, subsequently influenced stimuli. Importantly, relationship between hemodynamics as predicted our aligned well experimental analysis. HFD ACF had significant contributions displaying stability even presence high CO fluctuations. contrast, morphological features reduced contribution unstable conditions. Conclusion: Temporal essential single-site PPG-based estimation. Understanding physiological implications these can aid development algorithms clear interpretability optimal structures.

Language: Английский

---

Convolutional Autoencoder for Real-Time PPG Based Blood Pressure Monitoring Using TinyML

Published: Dec. 17, 2023

In this paper, we propose an efficient and robust convolutional autoencoder (CAE) model for continuous realtime blood pressure (BP) monitoring. The proposed was implemented on a resource-constrained edge device. built to capture the hidden patterns among successive segments alleviate effects of momentary glitches outliers. deployed assessed Arduino Nano 33 BLE Sense in real-time environment by means Tiny Machine Learning (TinyML). Extensive results revealed that improved BP prediction accuracy both offline experiments. With 4 features, achieved mean absolute error±standard deviation (MAE±SD) 2.81±2.84 1.51±1.85 mmHg systolic (SBP) diastolic (DBP), respectively, dataset 40 subjects. Whereas microcontroller unit (MCU) based predictions attained 2.25±2.82 SBP 5.01±2.10 DBP, 8 volunteers. Compared state-of-the-art models tiny devices, our showed superior robustness accuracy. Overall, study offered some important insights into significance compact impactful feature set effectiveness setting.

Language: Английский

---

Analysis of the Chaotic Component of Photoplethysmography and Its Association with Hemodynamic Parameters

Entropy, Journal Year: 2023, Volume and Issue: 25(12), P. 1582 - 1582

Published: Nov. 24, 2023

Wearable technologies face challenges due to signal instability, hindering their usage. Thus, it is crucial comprehend the connection between dynamic patterns in photoplethysmography (PPG) signals and cardiovascular health. In our study, we collected 401 multimodal recordings from two public databases, evaluating hemodynamic conditions like blood pressure (BP), cardiac output (CO), vascular compliance (C), peripheral resistance (R). Using irregular-resampling auto-spectral analysis (IRASA), quantified chaotic components PPG employed different methods measure fractal dimension (FD) entropy. Our findings revealed that surgery patients, power of increased with stiffness. As intensity CO fluctuations increased, there was a notable strengthening correlation most complexity measures these parameters. Interestingly, some conventional morphological features displayed significant decrease correlation, indicating shift static scenario. Healthy subjects exhibited higher percentage components, hemodynamics this group tended be more pronounced. Causal showed are main influencers for FD changes, observed feedback cases. conclusion, understanding vital assessing health, especially individuals unstable or during ambulatory testing. These insights can help overcome faced by wearable enhance usage real-world scenarios.

Language: Английский

---

Contribution of Timescale Patterns to Photoplethysmography-based Blood Pressure Estimation

Research Square (Research Square), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 6, 2023

Abstract Objective: Single-site photoplethysmography (PPG)-based blood pressure (BP) estimation has raised a lot of interest due to its compactness and low cost. However, this method relies on PPG morphological features, which are sensitive noise measurement conditions. The underlying physiological mechanism was also unclear at moment. In study, we propose add timescale patterns improve the BP performance clarify mechanism. Methods: In-silico simulation with four-element Windkessel model showed that peripheral resistance vascular compliance variation during cardiac cycle correlated PPG’s long- short-term self-similarity, significantly BP. A publicly available dataset used validate predictions using mutual information analysis regression assessment. Results: hemodynamic property cardiovascular system determines how fast responds stimulus. High or leads prolonged overlapped responses, could be described by patterns. Adding these increased PPG-BP improved performance. Compared algorithms biometric mean absolute error (MAE) calibrated systolic (SBP) reduced from 5.37mmHg 4.51mmHg, while MAE calibration-free diastolic (DBP) 3.46mmHg 2.81mmHg. median intra-subject correlation between SBP/DBP ground truth 0.63/0.34 0.80/0.68, means intrinsic fluctuation better captured. Conclusion: Timescale were vital single-site PPG-based estimation. Understanding implication may help us design clear interpretability simplified structures.

Language: Английский

---