Matti Kaisti

BackgroundHeart failure (HF) is the leading cause of hospitalization in individuals over 65 years of age. Identifying noninvasive methods to detect HF may address the epidemic of HF. Seismocardiography which measures cardiac vibrations transmitted to the chest wall has recently emerged as a promising technology to detect HF.ObjectivesIn this multicenter study, the authors examined whether seismocardiography using commercially available smartphones can differentiate control subjects from patients with stage C HF.MethodsBoth inpatients and outpatients with HF were enrolled from Finland and the United States. Inpatients with HF were assessed within 2 days of admission, and outpatients were assessed in the ambulatory setting. In a prespecified pooled data analysis, algorithms were derived using logistic regression and then validated using a bootstrap aggregation method.ResultsA total of 217 participants …

An apparatus for measuring compliance of blood vessels includes a photoplethysmography sensor for emitting electromagnetic radiation to the blood vessels, for receiving electromagnetic radiation reflected off the blood vessels, and for producing a measurement signal indicative of the received electromagnetic radiation. The apparatus further includes a pressure instrument for producing mechanical pressure applied on the blood vessels, and a control system for controlling the pressure instrument to change the mechanical pressure linearly with respect to time during emission of electromagnetic radiation to the blood vessels and reception of reflected electromagnetic radiation from the blood vessels. The control system finds, from the measurement signal, a portion whose envelope has exponential change with respect to time and produces an estimate for an exponent coefficient of time during the exponential …

Minimally invasive and non‐invasive hemodynamic monitoring technologies have recently gained more attention, driven by technological advances and the inherent risk of complications in invasive techniques. In this article, an experimental non‐invasive system is presented that effectively combines the capabilities of spectrometry, photoplethysmography (PPG), and arterial pressure measurement. Both time‐ and wavelength‐resolved optical signals from the fingertip are measured under external pressure, which gradually increased above the level of systolic blood pressure. The optical channels measured at 434–731 nm divided into three groups separated by a group of channels with wavelengths approximately between 590 and 630 nm. This group of channels, labeled transition band, is characterized by abrupt changes resulting from a decrease in the absorption coefficient of whole blood. External pressure …

Emotion recognition, a rapidly evolving domain in digital health, has witnessed significant transformations with the advent of personalized approaches and advanced machine learning (ML) techniques. These advancements have shifted the focus from traditional, generalized models to more individual-centric methodologies, underscoring the importance of understanding and catering to the unique emotional expressions of individuals. Our study delves into the concept of model personalization in emotion recognition, moving away from the one-size-fits-all approach. We conducted a series of experiments using the Emognition dataset, comprising physiological and video data of human subjects expressing various emotions, to investigate this personalized approach to affective computing. For the 10 individuals in the dataset with a sufficient representation of at least two ground truth emotion labels, we trained a personalized version of three classical ML models (k-nearest neighbors, random forests, and a dense neural network) on a set of 51 features extracted from each video frame. We ensured that all the frames used to train the models occurred earlier in the video than the frames used to test the model. We measured the importance of each facial feature for all the personalized models and observed differing ranked lists of the top features across the subjects, highlighting the need for model personalization. We then compared the personalized models against a generalized model trained using data from all 10 subjects. The mean F1 scores for the personalized models, specifically for the k-nearest neighbors, random forest, and dense neural …

The results and conclusions were, that altering the electrode material, distance, or size of the signal collection setup affects the resulting signal. The electrode material affects the signal quality the most, while the distance and the electrode size affect the average waveforms. The results presented in this thesis do not contradict any previous research on the same topic but do support their conclusions gathering the different alterations in the same reference framework.

Results from two independent clinical validation studies for measuring hemodynamics at the patient's bedside using a compact finger probe are reported. Technology comprises a barometric pressure sensor, and in one implementation, additionally, an optical sensor for photoplethysmography (PPG) is developed, which can be used to measure blood pressure and analyze rhythm, including the continuous detection of atrial fibrillation. The capabilities of the technology are shown in several form factors, including a miniaturized version resembling a common pulse oximeter to which the technology could be integrated in. Several main results are presented: i) the miniature finger probe meets the accuracy requirements of non‐invasive blood pressure instrument validation standard, ii) atrial fibrillation can be detected during the blood pressure measurement and in a continuous recording, iii) a unique comparison …

Traditionally, machine learning-based clinical prediction models have been trained and evaluated on patient data from a single source, such as a hospital. Cross-validation methods can be used to estimate the accuracy of such models on new patients originating from the same source, by repeated random splitting of the data. However, such estimates tend to be highly overoptimistic when compared to accuracy obtained from deploying models to sources not represented in the dataset, such as a new hospital. The increasing availability of multi-source medical datasets provides new opportunities for obtaining more comprehensive and realistic evaluations of expected accuracy through source-level cross-validation designs. In this study, we present a systematic empirical evaluation of standard K-fold cross-validation and leave-source-out cross-validation methods in a multi-source setting. We consider the task of electrocardiogram based cardiovascular disease classification, combining and harmonizing the openly available PhysioNet CinC Challenge 2021 and the Shandong Provincial Hospital datasets for our study. Our results show that K-fold cross-validation, both on single-source and multi-source data, systemically overestimates prediction performance when the end goal is to generalize to new sources. Leave-source-out cross-validation provides more reliable performance estimates, having close to zero bias though larger variability. The evaluation highlights the dangers of obtaining misleading cross-validation results on medical data and demonstrates how these issues can be mitigated when having access to multi-source data.

An apparatus for measuring a jugular vein pressure waveform includes a rotation sensor configured to produce a measurement signal when being against a skin of an individual and in a movement sensing relation with a jugular vein of the individual. The apparatus includes a processing system configured to receive the measurement signal and produce a waveform of a motion of the skin in a direction perpendicular to the skin based on the measurement signal indicative of rotation of the rotation sensor, where the waveform of the motion of the skin is indicative of the jugular vein pressure waveform. The rotation sensor that measures rotation is more insensitive to movements not related to variation of the jugular vein pressure than for example an acceleration sensor.

Strict privacy regulations pose challenges to the development of machine learning (ML) in the field of health technology where data is particularly sensitive. Gathering and using robust, bias-free, and suitably anonymized datasets required by ML models is difficult, time-consuming, and thus expensive. Parametric synthetic data offers a solution by mimicking real-world processes with easily adjustable parameters that shape the information content of the data as desired. This article presents a system demonstrating how synthetic data can be used in conjunction with wearable edge devices. Importantly, the system preserves privacy as there is no risk of leaking sensitive information from the model or during the use of the wearable device. The system consists of (1) a synthetic photoplethysmogram (PPG) model, (2) convolutional neural network (CNN) models trained with the synthetic signals, (3) a wearable edge …

The accurate quantification of hypersensitive response (HR) programmed cell death is imperative for understanding plant defense mechanisms and developing disease-resistant crop varieties. In this study, we report an accelerated phenotyping platform for the continuous-time, rapid and quantitative assessment of HR: Parallel Automated Spectroscopy Tool for Electrolyte Leakage (PASTEL). Compared to traditional HR assays, PASTEL significantly improves temporal resolution and has high sensitivity, facilitating the detection of microscopic levels of cell death. We validated PASTEL by transiently expressing the effector protein AVRblb2 in transgenic lines of the model plant Nicotiana benthamiana (expressing the corresponding resistance protein Rpi-blb2) to reliably induce HR. We were able to detect cell death at microscopic intensities, where leaf tissue appeared healthy to the naked eye one week after infiltration. PASTEL produces large amounts of frequency domain impedance data captured continuously (sub-seconds to minutes). Using this data, we developed a supervised machine learning models for classification of HR. We were able to classify input data (inclusive of our entire tested concentration range) as HR-positive or negative with 84.1% mean accuracy (F 1 score= 0.75) at 1 hour and with 87.8% mean accuracy (F 1 score= 0.81) at 22 hours. With PASTEL and the ML models produced in this work, it is possible to phenotype disease resistance in plants in hours instead of days to weeks.

Our aim is to develop a blood pressure (BP) measurement technology that could be integrated into a finger-worn pulse oximeter, eliminating the need for a brachial cuff. We present a miniature cuffless tonometric finger probe system that uses the oscillometric method to measure BP. Our approach uses a motorized press that is used to apply pressure to the fingertip to measure BP. We verified the functionality of the device in a clinical trial (n = 43) resulting in systolic and diastolic pressures ((mean ± SD) mmHg) of (−3.5 ± 8.4) mmHg and (−4.0 ± 4.4) mmHg, respectively. Comparison was made with manual auscultation (n = 26) and automated cuff oscillometry (n = 18). In addition to BP, we demonstrated the ability of the device to assess arterial stiffness (n = 18) and detect atrial fibrillation (n = 6). We were able to introduce a sufficiently small device that could be used for convenient ambulatory measurements with …

The right internal jugular vein is connected to the right atrium of the heart via the superior vena cava, and consequently its pressure, known as the jugular venous pressure or the jugular venous pulse (JVP), is an important indicator of cardiac function. The JVP can be estimated visually from the neck but it is rather difficult and imprecise. In this article we propose a method to measure the JVP using a motion sensor patch attached to the neck. The JVP signal was extracted from the sensor’s 3-axes gyroscope signal and aligned with simultaneously measured ECG and seismocardiogram signals.The method was tested on 20 healthy subjects. The timings of the characteristic JVP waves were compared with the ECG R peaks and seismocardiogram heart sounds S1 and S2. The JVP was reliably measured from 18 subjects with all three waves identified. The timings of the waves were also physiologically plausible when …

Mobile digital therapeutics for autism spectrum disorder (ASD) often target emotion recognition and evocation, which is a challenge for children with ASD. While such mobile applications often use computer vision machine learning (ML) models to guide the adaptive nature of the digital intervention, a single model is usually deployed and applied to all children. Here, we explore the potential of model personalization, or training a single emotion recognition model per person, to improve the performance of these underlying emotion recognition models used to guide digital health therapies for children with ASD. We conducted experiments on the Emognition dataset, a video dataset of human subjects evoking a series of emotions. For a subset of 10 individuals in the dataset with a sufficient representation of at least two ground truth emotion labels, we trained a personalized version of three classical ML models on a set of 51 features extracted from each video frame. We measured the importance of each facial feature for all personalized models and observed differing ranked lists of top features across subjects, motivating the need for model personalization. We then compared the personalized models against a generalized model trained using data from all 10 participants. The mean F1-scores achieved by the personalized models were 90.48%, 92.66%, and 86.40%, respectively. By contrast, the mean F1-scores reached by non-personalized models trained on different human subjects and evaluated using the same test set were 88.55%, 91.78%, and 80.42%, respectively. The personalized models outperformed the generalized models for 7 out of 10 …

Background Heart failure (HF) is the leading cause of hospitalization in people over the age of 65 years. More recently, cardiac motion sensor technology has emerged as a promising technology to detect HF. Methods In this multicenter study, we examined whether accelerometer and gyroscope signals from motion sensors collected using commercially available smartphones can classify HF status. Participants were enrolled from Finland and the United States. Participants hospitalized with acute decompensated HF were assessed in the acute state and re-assessed in the stabilized state prior to discharge. Outpatient participants were assessed in the stable state. In a pre-specified pooled data analysis, state specific algorithms to detect HF were first derived using logistic regression and validated using boostrap aggregation method (10 repeats) followed by sensitivity analysis in …

Microcirculation is a key compartment in the human cardiovascular system due to its vital roles in providing oxygen and nutrients to tissue, removing metabolic byproducts and regulating blood flow in organs. This network of small blood vessels, known as microvasculature, has been shown to have a link to many cardiovascular diseases (CVDs). This work presents a method capable of extracting information from different depths of the cutaneous vasculature, including the microvasculature, of the fingertip using an optical sensor with controllable external compression force. Our experiments show that the optical channels can be used to estimate blood pressure (BP) at different depths of the tissue, including shallow depths with microvascular blood vessels. Additionally, we show that shorter-wavelength optical signals (465, 515, and 590 nm) are more sensitive to pressure-induced vasodilation (PIV) than longer …

Continuous blood pressure (BP) monitoring gives a better understanding of a person’s cardiovascular health status than single BP measurements. The existing measurement techniques are often highly complex and expensive or suffer from inaccuracies. We propose a simple, yet effective technique for continuous BP monitoring. Our method is based on the finding that the nonpulsatile (dc) component of the photoplethysmograph (PPG) correlates with BP. By keeping the infrared (IR) PPG dc component constant by altering the applied external pressure using a feedback mechanism, the BP can be measured continuously. This way the pressure reading from the pressure sensor follows the mean intraarterial BP. We call this low-frequency vascular unloading. We propose a method for assessing the measurement error introduced by changes in vasomotor tone. Green PPG was used for the vasomotor compensation method. We packaged …

We present a method for training neural networks with synthetic electrocardiograms that mimic signals produced by a wearable single lead electrocardiogram monitor. We use domain randomization where the synthetic signal properties such as the waveform shape, RR-intervals and noise are varied for every training example. Models trained with synthetic data are compared to their counterparts trained with real data. Detection of r-waves in electrocardiograms recorded during different physical activities and in atrial fibrillation is used to assess the performance. By allowing the randomization of the synthetic signals to increase beyond what is typically observed in the real-world data the performance is on par or superseding the performance of networks trained with real data. Experiments show robust model performance using different seeds and on different unseen test sets that were fully separated from the training …

We developed a system for monitoring both activity and electrocardiogram for improved reliability of cardiac monitoring. Additionally, the link between activity information and recorded cardiac information can be used to better incorporate physiological state in analysis in free-living conditions. Our approach uses a machine learning model to predict the activity based on accelerometer data that is subsequently used to estimate cardiac monitoring reliability and linking the heart rate data in to a physical activity. We collected proof-of-concept data from eight healthy volunteers using accelerometers on wrist and on thigh and an electrocardiogram (ECG). The measurement protocol included eight activities (lying, sitting, standing, walking, jogging, walking stairs up and down and cycling). Each measurement was one minute long and the set was repeated 5-10 times per research participant. In addition, three individuals …

Differences in cardiac time intervals (CTIs) have previously been shown in different patient groups with varying levels of cardiac function. These studies relied on methods such as conventional echocardiography or tissue doppler imaging performed by a specialist to extract CTIs. The goal of this study was to evaluate the ability of using a combination of single lead ECG and 3-axis seismocardiography (SCG) from a sensor placed on a subject's sternum to automatically extract CTIs. For each subject, pre-ejection period (PEP), left ventricular ejection time ( $L$ VET), total systolic time $(TST)$ , and total diastolic time $(TDT)$ , which were normalized by the mean heart rate representing the entire recording were extracted using a custom developed algorithm. LVET was on average 20.5 % shorter in the NKHCD group $vs$ PRE-TAVI $(p< 0.05)$ ) and 5.9% shorter in the $HCD$ group $vs$ PRE-TAVI $(p> 0.05 …

Bed based ballistocardiography (BCG) is a prime candidate for at home and nighttime monitoring especially in the growing elderly population because co-operation from the user is not required to be able to record signals. One issue with BCG is that the signal quality has intra-and inter-person variability based on factors such as age, gender, body position, and motion artifacts, making it challenging to accurately measure heart rate. A rule-based algorithm which considers all eight available BCG channels simultaneously from a given time epoch was developed using continuous wavelet transform (CWT) to extract the localized time-frequency representation of each epoch and then an averaging method was applied across the different scales of the CWT to produce a 1-dimensional array. Autocorrelation was then applied to this array to produce a heart rate estimate based on the lag between the autocorrelation …