Machine and Deep Learning for Tuberculosis Detection on Chest X-Rays: Systematic Literature Review.

BACKGROUND: Tuberculosis (TB) was the leading infectious cause of mortality globally prior to COVID-19 and chest radiography has an important role in the detection, and subsequent diagnosis, of patients with this disease. The conventional experts reading has substantial within- and between-observer variability, indicating poor reliability of human readers. Substantial efforts have been made in utilizing various artificial intelligence-based algorithms to address the limitations of human reading of chest radiographs for diagnosing TB. OBJECTIVE: This systematic literature review (SLR) aims to assess the performance of machine learning (ML) and deep learning (DL) in the detection of TB using chest radiography (chest x-ray [CXR]). METHODS: In conducting and reporting the SLR, we followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. A total of 309 records were identified from Scopus, PubMed, and IEEE (Institute of Electrical and Electronics Engineers) databases. We independently screened, reviewed, and assessed all available records and included 47 studies that met the inclusion criteria in this SLR. We also performed the risk of bias assessment using Quality Assessment of Diagnostic Accuracy Studies version 2 (QUADAS-2) and meta-analysis of 10 included studies that provided confusion matrix results. RESULTS: Various CXR data sets have been used in the included studies, with 2 of the most popular ones being Montgomery County (n=29) and Shenzhen (n=36) data sets. DL (n=34) was more commonly used than ML (n=7) in the included studies. Most studies used human radiologist's report as the reference standard. Support vector machine (n=5), k-nearest neighbors (n=3), and random forest (n=2) were the most popular ML approaches. Meanwhile, convolutional neural networks were the most commonly used DL techniques, with the 4 most popular applications being ResNet-50 (n=11), VGG-16 (n=8), VGG-19 (n=7), and AlexNet (n=6). Four performance metrics were popularly used, namely, accuracy (n=35), area under the curve (AUC; n=34), sensitivity (n=27), and specificity (n=23). In terms of the performance results, ML showed higher accuracy (mean ~93.71%) and sensitivity (mean ~92.55%), while on average DL models achieved better AUC (mean ~92.12%) and specificity (mean ~91.54%). Based on data from 10 studies that provided confusion matrix results, we estimated the pooled sensitivity and specificity of ML and DL methods to be 0.9857 (95% CI 0.9477-1.00) and 0.9805 (95% CI 0.9255-1.00), respectively. From the risk of bias assessment, 17 studies were regarded as having unclear risks for the reference standard aspect and 6 studies were regarded as having unclear risks for the flow and timing aspect. Only 2 included studies had built applications based on the proposed solutions. CONCLUSIONS: Findings from this SLR confirm the high potential of both ML and DL for TB detection using CXR. Future studies need to pay a close attention on 2 aspects of risk of bias, namely, the reference standard and the flow and timing aspects. TRIAL REGISTRATION: PROSPERO CRD42021277155; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=277155.

Dynamic characteristics of oxygen consumption.

BACKGROUND: Previous studies have indicated that oxygen uptake ([Formula: see text]) is one of the most accurate indices for assessing the cardiorespiratory response to exercise. In most existing studies, the response of [Formula: see text] is often roughly modelled as a first-order system due to the inadequate stimulation and low signal to noise ratio. To overcome this difficulty, this paper proposes a novel nonparametric kernel-based method for the dynamic modelling of [Formula: see text] response to provide a more robust estimation. METHODS: Twenty healthy non-athlete participants conducted treadmill exercises with monotonous stimulation (e.g., single step function as input). During the exercise, [Formula: see text] was measured and recorded by a popular portable gas analyser ([Formula: see text], COSMED). Based on the recorded data, a kernel-based estimation method was proposed to perform the nonparametric modelling of [Formula: see text]. For the proposed method, a properly selected kernel can represent the prior modelling information to reduce the dependence of comprehensive stimulations. Furthermore, due to the special elastic net formed by [Formula: see text] norm and kernelised [Formula: see text] norm, the estimations are smooth and concise. Additionally, the finite impulse response based nonparametric model which estimated by the proposed method can optimally select the order and fit better in terms of goodness-of-fit comparing to classical methods. RESULTS: Several kernels were introduced for the kernel-based [Formula: see text] modelling method. The results clearly indicated that the stable spline (SS) kernel has the best performance for [Formula: see text] modelling. Particularly, based on the experimental data from 20 participants, the estimated response from the proposed method with SS kernel was significantly better than the results from the benchmark method [i.e., prediction error method (PEM)] ([Formula: see text] vs [Formula: see text]). CONCLUSIONS: The proposed nonparametric modelling method is an effective method for the estimation of the impulse response of VO2-Speed system. Furthermore, the identified average nonparametric model method can dynamically predict [Formula: see text] response with acceptable accuracy during treadmill exercise.

An equivalent circuit model for onset and offset exercise response.

BACKGROUND: The switching exercise (e.g., Interval Training) has been a commonly used exercise protocol nowadays for the enhancement of exerciser's cardiovascular fitness. The current difficulty for simulating human onset and offset exercise responses regarding the switching exercise is to ensure the continuity of the outputs during onset-offset switching, as well as to accommodate the exercise intensities at both onset and offset of exercise. METHODS: Twenty-one untrained healthy subjects performed treadmill trials following both single switching exercise (e.g., single-cycle square wave protocol) and repetitive switching exercise (e.g., interval training protocol). During exercise, heart rate (HR) and oxygen uptake (VO2) were monitored and recorded by a portable gas analyzer (K4b 2, Cosmed). An equivalent single-supply switching resistance-capacitor (RC) circuit model was proposed to accommodate the observed variations of the onset and offset dynamics. The single-cycle square wave protocol was utilized to investigate the respective dynamics at onset and offset of exercise with the aerobic zone of approximate 70%-77% of HR max, and verify the adaption feature for the accommodation of different exercise strengths. The design of the interval training protocol was to verify the transient properties during onset-offset switching. A verification method including Root-mean-square-error (RMSE) and correlation coefficient, was introduced for comparisons between the measured data and model outputs. RESULTS: The experimental results from single-cycle square wave exercises clearly confirm that the onset and offset characteristics for both HR and VO2 are distinctly different. Based on the experimental data for both single and repetitive square wave exercise protocols, the proposed model was then presented to simulate the onset and offset exercise responses, which were well correlated indicating good agreement with observations. CONCLUSIONS: Compared with existing works, this model can accommodate the different exercise strengths at both onset and offset of exercise, while also depicting human onset and offset exercise responses, and guarantee the continuity of outputs during onset-offset switching. A unique adaption feature by allowing the time constant and steady state gain to re-shift back to their original states, more closely mimics the different exercise strengths during normal daily exercise activities.

Design of a multi-site multi-state clinical trial of home monitoring of chronic disease in the community in Australia.

BACKGROUND: Telehealth services based on at-home monitoring of vital signs and the administration of clinical questionnaires are being increasingly used to manage chronic disease in the community, but few statistically robust studies are available in Australia to evaluate a wide range of health and socio-economic outcomes. The objectives of this study are to use robust statistical methods to research the impact of at home telemonitoring on health care outcomes, acceptability of telemonitoring to patients, carers and clinicians and to identify workplace cultural factors and capacity for organisational change management that will impact on large scale national deployment of telehealth services. Additionally, to develop advanced modelling and data analytics tools to risk stratify patients on a daily basis to automatically identify exacerbations of their chronic conditions. METHODS/DESIGN: A clinical trial is proposed at five locations in five states and territories along the Eastern Seaboard of Australia. Each site will have 25 Test patients and 50 case matched control patients. All participants will be selected based on clinical criteria of at least two hospitalisations in the previous year or four or more admissions over the last five years for a range of one or more chronic conditions. Control patients are matched according to age, sex, major diagnosis and their Socio-Economic Indexes for Areas (SEIFA). The Trial Design is an Intervention control study based on the Before-After-Control-Impact (BACI) design. DISCUSSION: Our preliminary data indicates that most outcome variables before and after the intervention are not stationary, and accordingly we model this behaviour using linear mixed-effects (lme) models which can flexibly model within-group correlation often present in longitudinal data with repeated measures. We expect reduced incidence of unscheduled hospitalisation as well as improvement in the management of chronically ill patients, leading to better and more cost effective care. Advanced data analytics together with clinical decision support will allow telehealth to be deployed in very large numbers nationally without placing an excessive workload on the monitoring facility or the patient's own clinicians. TRIAL REGISTRATION: Registered with Australian New Zealand Clinical Trial Registry on 1st April 2013. Trial ID: ACTRN12613000635763.

Measuring blood pressure from Korotkoff sounds as the brachial cuff inflates on average provides higher values than when the cuff deflates.

Objectives. In this study, we test the hypothesis that if, as demonstrated in a previous study, brachial arteries exhibit hysteresis as the occluding cuff is deflated and fail to open until cuff pressure (CP) is well below true intra-arterial blood pressure (IAPB), estimating systolic (SBP) and diastolic blood pressure (DBP) from the presence of Korotkoff sounds (KS) as CP increases may eliminate these errors and give more accurate estimates of SBP and DBP relative to IABP readings.Approach. In 62 subjects of varying ages (45.1 ± 19.8, range 20.6-75.8 years), including 44 men (45.3 ± 19.4, range 20.6-75.8 years) and 18 women (44.4 ± 21.4, range 20.9-75.3 years), we sequentially recorded SBP and DBP both during cuff inflation and cuff deflation using KS.Results. There was a significant (p< 0.0001) increase in SBP from 122.8 ± 13.2 to 127.6 ± 13.0 mmHg and a significant (p= 0.0001) increase in DBP from 70.0 ± 9.0 to 77.5 ± 9.7 mmHg. Of the 62 subjects, 51 showed a positive increase in SBP (0-14 mmHg) and 11 subjects showed a reduction (-0.3 to -7 mmHg). The average differences for SBP and DBP estimates derived as the cuff inflates and those derived as the cuff deflates were 4.8 ± 4.6 mmHg and 2.5 ± 4.6 mmHg, not dissimilar to the differences reported between IABP and non-invasive blood pressure measurements. Although we could not develop multiparameter linear or non-linear models to explain this phenomenon we have clearly demonstrated through ANOVA tests that both body mass index (BMI) and pulse wave velocity are implicated, supporting the hypothesis that the phenomenon is associated with age, higher BMI and stiffer arteries.Significance. The implications of this study are that brachial sphygmomanometry carried out during cuff inflation could be more accurate than measurements carried out as the cuff deflates. Further research is required to validate these results with IAPB measurements.

Accurate detection of Korotkoff sounds reveals large discrepancy between intra-arterial systolic pressure and simultaneous noninvasive measurement of blood pressure with brachial cuff sphygmomanometry.

Cardiovascular disease is the number 1 cause of death globally, with elevated blood pressure (BP) being the single largest risk factor. Hence, BP is an important physiological parameter used as an indicator of cardiovascular health. Noninvasive cuff-based automated monitoring is now the dominant method for BP measurement and irrespective of whether the oscillometric or the auscultatory method is used, all are calibrated according to the Universal Standard (ISO 81060-2:2019), which requires two trained operators to listen to Korotkoff K1 sounds for SBP and K4/K5 sounds for DBP. Hence, Korotkoff sounds are fundamental to the calibration of all NIBP devices. In this study of 40 lightly sedated patients, aged 64.1 ±â€Š9.6 years, we compare SBP and DBP recorded directly by intra-arterial fluid filled catheters to values recorded from the onset (SBP-K) and cessation (DBP-K) of Korotkoff sounds. We demonstrate that whilst DBP-K measurements are in good agreement, with a mean difference of -0.3 ±â€Š5.2 mmHg, SBP-K underestimates true intra-arterial SBP (IA-SBP) by an average of 14 ±â€Š9.6 mmHg. The underestimation arises from delays in the re-opening of the brachial artery following deflation of the brachial cuff to below SBP. The reasons for this delay are not known but appear related to the difference between SBP and the pressure under the cuff as blood first begins to flow, as the cuff deflates. Linear models are presented that can correct the underestimation in SBP resulting in estimates with a mean difference of 0.2 ±â€Š7.1 mmHg with respect to intra-arterial SBP.

Validation of oscillometric ratio and maximum gradient methods for non-invasive blood pressure measurement with intra-arterial blood pressure measurements as reference.

Most non-invasive blood pressure (BP) measurements are carried out using instruments which implement either the Ratio or the Maximum Gradient oscillometric method, mostly during cuff deflation, but more rarely during cuff inflation. Yet, there is little published literature on the relative advantages and accuracy of these two methods. In this study of 40 lightly sedated individuals aged 64.1 ± 9.6 years, we evaluate and compare the performance of the oscillometric ratio (K) and gradient (Grad) methods for the non-invasive estimation of mean pressure, SBP and DBP with reference to invasive intra-arterial values. There was no significant difference between intra-arterial estimates of mean pressure made via Korotkoff sounds (MP-OWE) or the gradient method (MP-Grad). However, 17.7% of MP-OWE and 15% of MP-Grad were in error by more than 10 mmHg. SBP-K and SBP-Grad underestimated SBP by 14 and 18 mmHg, whilst accurately estimating DBP with mean errors of 0.4 ±â€Š5.0 and 1.7 ±â€Š6.1 mmHg, respectively. Relative to the reference standard SBP-K, SBP-Grad and DBP-Grad were estimated with a mean error of -4.5 ±â€Š6.6 and 1.4 ±â€Š5.6 mmHg, respectively, noting that using the full range of recommended ratios introduces errors of 12 and 7 mmHg in SBP and DBP, respectively. We also show that it is possible to find ratios which minimize the root mean square error (RMSE) and the mean error for any particular individual cohort. We developed linear models for estimating SBP and SBP-K from a range of demographic and non-invasive OWE variables with resulting mean errors of 0.15 ±â€Š5.6 and 0.3 ±â€Š5.7 mmHg, acceptable according to the Universal standard.

Comparison of cuff inflation and cuff deflation brachial sphygmomanometry with intra-arterial blood pressure as reference.

Conventional sphygmomanometry with cuff deflation is used to calibrate all noninvasive BP (NIBP) instruments and the International Standard makes no mention of calibrating methods specifically for NIBP instruments, which estimate systolic and diastolic pressure during cuff inflation rather than cuff deflation. There is however increasing interest in inflation-based NIBP (iNIBP) instruments on the basis of shorter measurement time, reduction in maximal inflation pressure and improvement in patient comfort and outcomes. However, we have previously demonstrated that SBP estimates based on the occurrence of the first K1 Korotkoff sounds during cuff deflation can underestimate intra-arterial SBP (IA-SBP) by an average of 14 ±â€Š10 mmHg. In this study, we compare the dynamics of intra-arterial blood pressure (IABP) measurements with sequential measurement of Korotkoff sounds during both cuff inflation and cuff deflation in the same individual. In 40 individuals aged 64.1 ±â€Š9.6 years (range 36-86 years), the overall dynamic responses below the cuff were similar, but the underestimation error was significantly larger during inflation than deflation, increasing from 14 ±â€Š10 to 19 ±â€Š12 mmHg ( P  < 0.0001). No statistical models were found which could compensate for this error as were found for cuff deflation. The statistically significant BP differences between inflation and deflation protocols reported in this study suggest different behaviour of the arterial and venous vasculature between arterial opening and closing which warrant further investigation, particularly for iNIBP devices reporting estimates during cuff inflation. In addition, measuring Korotkoff sounds during cuff inflation represents significant technical difficulties because of increasing pump motor noise.

Evaluation of the oscillometric method for noninvasive blood pressure measurement during cuff deflation and cuff inflation with reference to intra-arterial blood pressure.

There is little quantitative clinical data available to support blood pressure measurement accuracy during cuff inflation. In this study of 35 male and 5 female lightly anaesthetized subjects aged 64.1 ±â€Š9.6 years, we evaluate and compare the performance of both the oscillometric ratio and gradient methods during cuff deflation and cuff inflation with reference to intra-arterial measurements. We show that the oscillometric waveform envelopes (OWE), which are key to both methods, exhibit significant variability in both shape and smoothness leading to at least 15% error in the determination of mean pressure (MP). We confirm the observation from our previous studies that K1 Korotkoff sounds underestimate systolic blood pressure (SBP) and note that this underestimation is increased during cuff inflation. The estimation of diastolic blood pressure (DBP) is generally accurate for both the ratio and the gradient method, with the latter showing a significant increase during inflation. Since the gradient method estimates SBP and DBP from points of maximum gradient on each OWE recorded, it may offer significant benefits over the ratio method. However, we have shown that the ratio method can be optimized for any data set to achieve either a minimum mean error (ME) of close to 0 mmHg or minimum root mean square error (RMSE) with standard deviation (SD) of <5.0 mmHg. We conclude that whilst cuff inflation may offer some advantages, these are neither significant nor substantial, leaving as the only benefit, the potential for more rapid measurement and less patient discomfort.

A Novel Deep Ensemble Method for Selective Classification of Electrocardiograms.

OBJECTIVE: Telehealth paradigms are essential for remotely managing patients with chronic conditions. To assist clinicians in handling the large volumes of data collected through these systems, clinical decision support systems (CDSSs) have been developed. However, the effectiveness of CDSSs depends on the quality of remotely recorded physiological data and the reliability of the algorithms used for processing this data. This study aims to reliably detect atrial fibrillation (AF) from short-term single-lead (STSL) electrocardiogram (ECG) recordings obtained in unsupervised telehealth environments. METHODS: A novel deep ensemble-based method was developed for detecting AF from STSL ECG recordings. Following this, a postprocessing algorithm was created to assess uncertainty in classified STSL ECGs and to refrain from interpretation when confidence is low. The proposed method was validated through a 5-fold cross-validation on the Cardiology Challenge 2017 (CinC2017) dataset. RESULTS: The deep ensemble method achieved 83.5 ± 1.5% sensitivity, 98.4 ± 0.2% specificity, and an F 1-score of 0.847 ± 0.016in AF detection. Implementing the selective classification algorithm resulted in significant improvements, with sensitivity increasing to 92.8 ± 2.2%, specificity to 99.7 ± 0.0%, and an F 1-score of 0.919 ± 0.016. CONCLUSION: The proposed method demonstrates the feasibility of accurately detecting AF from STSL ECG recordings. The selective classification approach offers a substantial enhancement to automated ECG interpretation algorithms in telehealth solutions. SIGNIFICANCE: These findings highlight the potential for improving the utility of telehealth systems by integrating advanced CDSSs capable of managing uncertainty and ensuring higher accuracy, thereby improving patient outcomes in remote healthcare settings.

New Perspectives on Non-invasive Blood Pressure Measurement.

Noninvasive blood pressure (NIBP) devices are calibrated against validated auscultation sphygmomanometers using Korotkoff sounds. This study aimed to investigate the timing of Korotkoff sounds in relation to pulse appearance in the brachial artery and values of intra-arterial blood pressure. Experiments were carried out on 15 participants, (14 males, 64.3 ± 10.4 years; one female, 86 yo), undergoing coronary angiography. A conventional occluding cuff, with a microphone for Korotkoff sounds, was placed on the upper arm (on the brachial artery). Intra-arterial blood pressure (IABP) was measured below the cuff with a fluid-filled catheter inserted via the radial artery and an external transducer. Finger photoplethysmography was used to measure brachial pulse wave velocity (PWV). Korotkoff sounds were processed electronically and custom algorithms identified the cuff pressure (CP) at which the first and last Korotkoff sounds were heard. PWV and max slope of the IABP pressure pulse were recorded to estimate arterial stiffness. The brachial artery closed at a CP of 132.0 ± 17.1 mmHg. Systolic and diastolic blood pressure (SBP and DBP) were 147.6 ± 14.3 and 72.7 ± 10.1 mmHg; mean pressure (MP, 100.1 ± 10.4 mmHg) was similar to MP derived from the peak of the oscillogram (98.5 ± 13.6 mmHg). Difference between IABP and CP recorded at first and last occurrence of Korotkoff sounds were, SBP: 19.0 ± 8.3 (range 2-29) mmHg, DBP: 4.0 ± 4.3 (range 2-12) mmHg. SBP derived from the onset of Korotkoff sounds can underestimate IABP by up to 19 mmHg. Since Korotkoff sounds are the recommended method mandated by the universal standard for the validation of blood pressure measuring devices, these errors are propagated through to all NIBP measurement devices irrespective of whether they use auscultatory or oscillometric methods.

Revisiting Transfer Learning Method for Tuberculosis Diagnosis.

Transfer learning (TL) has been proven to be a good strategy for solving domain-specific problems in many deep learning (DL) applications. Typically, in TL, a pre-trained DL model is used as a feature extractor and the extracted features are then fed to a newly trained classifier as the model head. In this study, we propose a new ensemble approach of transfer learning that uses multiple neural network classifiers at once in the model head. We compared the classification results of the proposed ensemble approach with the direct approach of several popular models, namely VGG-16, ResNet-50, and MobileNet, on two publicly available tuberculosis datasets, i.e., Montgomery County (MC) and Shenzhen (SZ) datasets. Moreover, we also compared the results when a fully pre-trained DL model was used for feature extraction versus the cases in which the features were obtained from a middle layer of the pre-trained DL model. Several metrics derived from confusion matrix results were used, namely the accuracy (ACC), sensitivity (SNS), specificity (SPC), precision (PRC), and F1-score. We concluded that the proposed ensemble approach outperformed the direct approach. Best result was achieved by ResNet-50 when the features were extracted from a middle layer with an accuracy of 91.2698% on MC dataset.Clinical Relevance- The proposed ensemble approach could increase the detection accuracy of 7-8% for Montgomery County dataset and 4-5% for Shenzhen dataset.

Assessing the Generalizability of a Deep Learning-based Automated Atrial Fibrillation Algorithm.

Automated detection of atrial fibrillation (AF) from electrocardiogram (ECG) traces remains a challenging task and is crucial for telemonitoring of patients after stroke. This study aimed to quantify the generalizability of a deep learning (DL)-based automated ECG classification algorithm. We first developed a novel hybrid DL (HDL) model using the PhysioNet/CinC Challenge 2017 (CinC2017) dataset (publicly available) that can classify the ECG recordings as one of four classes: normal sinus rhythm (NSR), AF, other rhythms (OR), and too noisy (TN) recordings. The (pre)trained HDL was then used to classify 636 ECG samples collected by our research team using a handheld ECG device, CONTEC PM10 Portable ECG Monitor, from 102 (age: 68 ± 15 years, 74 male) outpatients of the Eastern Heart Clinic and inpatients in the Cardiology ward of Prince of Wales Hospital, Sydney, Australia. The proposed HDL model achieved average test F1-score of 0.892 for NSR, AF, and OR, relative to the reference values, on the CinC2017 dataset. The HDL model also achieved an average F1-score of 0.722 (AF: 0.905, NSR: 0.791, OR: 0.471 and TN: 0.342) on the dataset created by our research team. After retraining the HDL model on this dataset using a 5-fold cross validation method, the average F1-score increased to 0.961. We finally conclude that the generalizability of the HDL-based algorithm developed for AF detection from short-term single-lead ECG traces is acceptable. However, the accuracy of the pre-trained DL model was significantly improved by retraining the model parameters on the new dataset of ECG traces.

Unsupervised machine-learning method for improving the performance of ambulatory fall-detection systems.

BACKGROUND: Falls can cause trauma, disability and death among older people. Ambulatory accelerometer devices are currently capable of detecting falls in a controlled environment. However, research suggests that most current approaches can tend to have insufficient sensitivity and specificity in non-laboratory environments, in part because impacts can be experienced as part of ordinary daily living activities. METHOD: We used a waist-worn wireless tri-axial accelerometer combined with digital signal processing, clustering and neural network classifiers. The method includes the application of Discrete Wavelet Transform, Regrouping Particle Swarm Optimization, Gaussian Distribution of Clustered Knowledge and an ensemble of classifiers including a multilayer perceptron and Augmented Radial Basis Function (ARBF) neural networks. RESULTS: Preliminary testing with 8 healthy individuals in a home environment yields 98.6% sensitivity to falls and 99.6% specificity for routine Activities of Daily Living (ADL) data. Single ARB and MLP classifiers were compared with a combined classifier. The combined classifier offers the greatest sensitivity, with a slight reduction in specificity for routine ADL and an increased specificity for exercise activities. In preliminary tests, the approach achieves 100% sensitivity on in-group falls, 97.65% on out-group falls, 99.33% specificity on routine ADL, and 96.59% specificity on exercise ADL. CONCLUSION: The pre-processing and feature-extraction steps appear to simplify the signal while successfully extracting the essential features that are required to characterize a fall. The results suggest this combination of classifiers can perform better than MLP alone. Preliminary testing suggests these methods may be useful for researchers who are attempting to improve the performance of ambulatory fall-detection systems.

Measuring blood pressure from Korotkoff sounds as the brachial cuff inflates on average provides higher values than when the cuff deflates.

Objective. In this study, we test the hypothesis that if, as demonstrated in a previous study, brachial arteries exhibit hysteresis as the occluding cuff is deflated and fail to open until cuff pressure (CP) is well below true intra-arterial blood pressure (IABP). Approach Estimating systolic (SBP) and diastolic blood pressure (DBP) from the presence of Korotkoff sounds as CPincreasesmay eliminate these errors and give more accurate estimates of SBP relative to IABP readings.Main Results.In 63 subjects of varying age 45.4 ± 19.9 years (range 21-76 years), including 44 men (45.2 ± 19.5, range 21-76 years) and 19 women (45.6 ± 21.4, range 21-75 years), there was a significant (p< 0.0001) increase in SBP from 124.4 ± 15.7 to 129.2 ± 16.3 mmHg and a significant (p< 0.0001) increase in DBP from 70.2 ± 10.7 to 73.6 ± 11.5 mmHg. Of the 63 subjects, 59 showed a positive increase in SBP (1-19 mmHg) and 5 subjects showed a reduction (-5 to -1 mmHg). The average differences for SBP estimates derived as the cuff inflates and estimates derived as the cuff deflates were 4.9 ± 4.7 mmHg, not dissimilar to the differences observed between IABP and NIBP measurements. Although we could not develop multiparameter linear or nonlinear models to explain this phenomenon we have clearly demonstrated through analysis of variance test that both body mass index (BMI) and pulse wave velocity are implicated, supporting the hypothesis that the phenomenon is associated with age, higher BMI and stiffer arteries.Significance. The implications of this study are potentially profound requiring the implementation of a new paradigm for NIBP measurement and a revision of the international standards for their calibration.

Artificial Intelligence Based Blood Pressure Estimation From Auscultatory and Oscillometric Waveforms: A Methodological Review.

Cardiovascular disease is known as the number one cause of death globally, with elevated blood pressure (BP) being the single largest risk factor. Hence, BP is an important physiological parameter used as an indicator of cardiovascular health. The use of automated non-invasive blood pressure (NIBP) measurement devices is growing, as they can be used without expertise and BP measurement can be performed by patients at home. Non-invasive cuff-based monitoring is the dominant method for BP measurement. While the oscillometric technique is most common, some automated NIBP measurement methods have been developed based on the auscultatory technique. By utilizing (relatively) large BP data annotated by experts, models can be trained using machine learning and statistical concepts to develop novel NIBP estimation algorithms. Amongst artificial intelligence (AI) techniques, deep learning has received increasing attention in different fields due to its strength in data classification and feature extraction problems. This paper reviews AI-based BP estimation methods with a focus on recent advances in deep learning-based approaches within the field. Various architectures and methodologies proposed todate are discussed to clarify their strengths and weaknesses. Based on the literature reviewed, deep learning brings plausible benefits to the field of BP estimation. We also discuss some limitations which can hinder the widespread adoption of deep learning in the field and suggest frameworks to overcome these challenges.

A Comparison of Haemodynamic Responses between Head-Up Tilt and Lower Body Negative Pressure.

Orthostatic intolerance (OI), a disorder of the autonomic nervous system, it is the development of symptoms when standing upright which are relieved when reclining. Head-up tilt (HUT) table test is a common test for assessing orthostatic tolerance. However, HUT is limited with low sensitivity and specificity. Another approach to stimulate the changing direction and value of the gravity field vector is the lower body negative pressure (LBNP) chamber. The aims of the study is to evaluate the physiological responses of healthy subjects on HUT and LBNP, and examine the relations of two tests. A total of 19 subjects were recruited. A validated wearable device, Sotera Visi Mobile was use to collect physiological signals simultaneously throughout the experiment procedures. Each subject went through a baseline supine rest, 70o of HUT test, another round of baseline supine rest, followed by activation of LBNP test. Three level of suction were applied, i.e. -30 mmHg, -40 mmHg, and -50 mmHg. In this pilot study, healthy subjects showed significantly increased of heart rate, and decreased of systolic blood pressure and diastolic blood pressure, in both HUT and LBNP tests. Although both tests are capable of stimulating a decreased blood volume in the central circulation, but the physiological responses behaved differently and shown only very week correlation. This suggesting that a combination of LBNP test with HUT test might work the best in orthostatic intolerance assessment.

A Novel Automated Blood Pressure Estimation Algorithm Using Sequences of Korotkoff Sounds.

The use of automated non-invasive blood pressure (NIBP) measurement devices is growing, as they can be used without expertise, and BP measurement can be performed by patients at home. Non-invasive cuff-based monitoring is the dominant method for BP measurement. While the oscillometric technique is most common, a few automated NIBP measurement methods have been developed based on the auscultatory technique. Amongst artificial intelligence (AI) techniques, deep learning has received increasing attention in different fields due to its strength in data classification, and feature extraction problems. This paper proposes a novel automated AI-based technique for NIBP estimation from auscultatory waveforms (AWs) based on converting the NIBP estimation problem to a sequence-to-sequence classification problem. To do this, a sequence of segments was first formed by segmenting the AWs, and their corresponding decomposed detail, and approximation parts obtained by wavelet packet decomposition method, and extracting features from each segment. Then, a label was assigned to each segment, i.e. (i) between systolic, and diastolic segments, and (ii) otherwise, and a bidirectional long short term memory recurrent neural network (BiLSTM-RNN) was devised to solve the resulting sequence-to-sequence classification problem. Adopting a 5-fold cross-validation scheme, and using a data base of 350 NIBP recordings gave an average mean absolute error of 1.7±3.7 mmHg for systolic BP (SBP), and 3.4 ±5.0 mmHg for diastolic BP (DBP) relative to reference values. Based on the results achieved, and comparisons made with the existing literature, it is concluded that the proposed automated BP estimation algorithm based on deep learning methods, and auscultatory waveform brings plausible benefits to the field of BP estimation.

Human Motion Intent Description Based on Bumpless Switching Mechanism for Rehabilitation Robot.

This paper aims to improve the performance of an electromyography (EMG) decoder based on a switching mechanism in controlling a rehabilitation robot for assisting human-robot cooperation arm movements. For a complex arm movement, the major difficulty of the EMG decoder modeling is to decode EMG signals with high accuracy in real-time. Our recent study presented a switching mechanism for carving up a complex task into simple subtasks and trained different submodels with low nonlinearity. However, it was observed that a "bump" behavior of decoder output (i.e., the discontinuity) occurred during the switching between two submodels. The bumps might cause unexpected impacts on the affected limb and thus potentially injure patients. To improve this undesired transient behavior on decoder outputs, we attempt to maintain the continuity of the outputs during the switching between multiple submodels. A bumpless switching mechanism is proposed by parameterizing submodels with all shared states and applied in the construction of the EMG decoder. Numerical simulation and real-time experiments demonstrated that the bumpless decoder shows high estimation accuracy in both offline and online EMG decoding. Furthermore, the outputs achieved by the proposed bumpless decoder in both testing and verification phases are significantly smoother than the ones obtained by a multimodel decoder without a bumpless switching mechanism. Therefore, the bumpless switching approach can be used to provide a smooth and accurate motion intent prediction from multi-channel EMG signals. Indeed, the method can actually prevent participants from being exposed to the risk of unpredictable loads.

Continuous Description of Human 3D Motion Intent Through Switching Mechanism.

Post-stroke motor recovery highly relies on voluntarily participating in active rehabilitation as early as possible for promoting the reorganization of the patient's brain. In this paper, a new method is proposed which manipulates cable-based rehabilitation robots to assist multi-joint body motions. This uses an electromyography (EMG) decoder for continuous estimation of voluntary motion intention to establish a cooperative human-machine interface for promoting the participation in rehabilitation exercises. In particular, for multi-joint complex tasks in three-dimensional space, a switching mechanism has been developed which can carve up tasks into separate simple motions. For each simple motion, a linear six-inputs and three-outputs time-invariant model is established respectively. The inputs are the processed muscle activations of six arm muscles, and the outputs are voluntary forces of participants when executing a multi-directional tracking task with visual feedback. The experiments for examining the decoder model and EMG-based controller include model training, testing and controller application phases with seven healthy participants. Experimental results demonstrate that the decoder model with the switching mechanism could effectively recognize arm movement intention and provide appropriate assistance to the participants. This study finds that the switching mechanism can improve both the model estimation accuracy and the completeness for executing complex tasks.