A framework for comparative study of databases and computational methods for arrhythmia detection from single-lead ECG.

Arrhythmia detection from ECG is an important area of computational ECG analysis. However, although a large number of public ECG recordings are available, most research uses only few datasets, making it difficult to estimate the generalizability of the plethora of ECG classification methods. Furthermore, there is a large variability in the evaluation procedures, as well as lack of insight into whether they could successfully perform in a real-world setup. To address these problems, we propose an open-source, flexible and configurable ECG classification codebase-ECGDL, as one of the first efforts that includes 9 arrhythmia datasets, covering a large number of both morphological and rhythmic arrhythmias, as well as 4 deep neural networks, 4 segmentation techniques and 4 evaluation schemes. We perform a comparative analysis along these framework components to provide a comprehensive perspective into arrhythmia classification, focusing on single-lead ECG as the most recent trend in wireless ECG monitoring. ECGDL unifies the class information representation in datasets by creating a label dictionary. Furthermore, it includes a set of the best-performing deep learning approaches with varying signal segmentation techniques and network architectures. A novel evaluation scheme, inter-patient cross-validation, has also been proposed to perform fair evaluation and comparison of results.

Power Autonomy Estimation of Low-Power Sensor for Long-Term ECG Monitoring.

The paper analyses the autonomy of a wireless body sensor that continuously measures the potential difference between two proximal electrodes on the skin, primarily used for measuring an electrocardiogram (ECG) when worn on the torso. The sensor is powered by a small rechargeable battery and is designed for extremely low power use. However, the autonomy of the sensor, regarding its power consumption, depends significantly on the measurement quality selection, which directly influences the amount of data transferred. Therefore, we perform an in-depth analysis of the power consumption sources, particularly those connected with the Bluetooth Low Energy (BLE) communication protocol, in order to model and then tune the autonomy of the wireless low-power body sensor for long-term ECG monitoring. Based on the findings, we propose two analytical models for power consumption: one for power consumption estimation in idle mode and the other one for power estimation in active mode. The proposed models are validated with the measured power consumption of the ECG sensor at different ECG sensor settings, such as sampling rate and transmit power. The proposed models show a good fit to the measured power consumption at different ECG sensor sampling rates. This allows for power consumption analysis and sensor autonomy predictions for different sensor settings. Moreover, the results show that the transmit power has a negligible effect on the sensor autonomy in the case of streaming data with high sampling rates. The most energy can be saved by lowering the sampling rate with suitable connection interval and by packing as much data as possible in a single BLE packet.

Patient-Specific Heartbeat Classification in Single-Lead ECG using Convolutional Neural Network.

For an expert cardiologist, any abnormality in the heart rhythm or electrocardiogram (ECG) shape can be easily detected as a sign of arrhythmia. However, this is a big challenge for a computer system. The need for automatic arrhythmia recognition comes from the development of many portable ECG measuring devices designed to function as a part of health monitoring platforms. These platforms, because of their wide availability, generate a lot of data and hence the need for algorithms to process this data. From the many methods for automatic heartbeat classification, convolutional neural networks (CNNs) are increasingly being applied in this ECG analysis task. The purpose of this paper is to develop arrhythmia classification model according to the standards defined by the Association for the Advancement of Medical Instruments (AAMI), using CNNs, on data from the publicly available MIT-BIH Arrhythmia database. We experiment with two types of heartbeat segmentation: static and dynamic. The ultimate goal is to implement an algorithm for long-term monitoring of a user's health, which is why we have focused on classification models from single-lead ECG, and, even more, on algorithms specifically designed for one person rather than general models. Therefore, we evaluate patient-specific CNN models also on measurements from a novel wireless single-lead ECG sensor.

Framework for Personalized Real-Time Control of Hidden Temperature Variables in Therapeutic Knee Cooling.

The paper formalizes, implements and evaluates a framework for personalized real-time control of inner knee temperature during cryotherapy after knee surgery. Studies have shown that the cryotherapy should be controlled depending on the individual patient's feedback on the cooling, which raises the need for smart personalized therapy. The framework is based on the feedback control loop that uses predicted instead of measured inner temperatures because measurements are not feasible or would introduce invasiveness into the system. It uses machine learning to construct a predictive model for estimation of the controlled inner temperature variable based on other variables whose measurement is more feasible - temperatures on the body surface. The machine learning method uses data generated from computer simulation of the therapeutic treatment for different input simulation parameters. A fuzzy proportional-derivative controller is designed to provide adequate near real-time control of the inner knee temperature by controlling the cooling temperature. The framework is evaluated for robustness and controllability. The results show that controlled cooling is essential for small-sized (and large-sized) knees that are significantly more (less) sensitive to the cooling compared to average-sized knees. Moreover, the framework recognizes dynamic physiological changes and potential changes in the system settings, such as extreme changes in the blood flow or changed target inner knee temperature, and consequently adapts the cooling temperature to reach the target value.

Medical-Grade ECG Sensor for Long-Term Monitoring.

The recent trend in electrocardiogram (ECG) device development is towards wireless body sensors applied for patient monitoring. The ultimate goal is to develop a multi-functional body sensor that will provide synchronized vital bio-signs of the monitored user. In this paper, we present an ECG sensor for long-term monitoring, which measures the surface potential difference between proximal electrodes near the heart, called differential ECG lead or differential lead, in short. The sensor has been certified as a class IIa medical device and is available on the market under the trademark Savvy ECG. An improvement from the user's perspective-immediate access to the measured data-is also implemented into the design. With appropriate placement of the device on the chest, a very clear distinction of all electrocardiographic waves can be achieved, allowing for ECG recording of high quality, sufficient for medical analysis. Experimental results that elucidate the measurements from a differential lead regarding sensors' position, the impact of artifacts, and potential diagnostic value, are shown. We demonstrate the sensors' potential by presenting results from its various areas of application: medicine, sports, veterinary, and some new fields of investigation, like hearth rate variability biofeedback assessment and biometric authentication.

Non-invasive real-time prediction of inner knee temperatures during therapeutic cooling.

The paper addresses the issue of non-invasive real-time prediction of hidden inner body temperature variables during therapeutic cooling or heating and proposes a solution that uses computer simulations and machine learning. The proposed approach is applied on a real-world problem in the domain of biomedicine - prediction of inner knee temperatures during therapeutic cooling (cryotherapy) after anterior cruciate ligament (ACL) reconstructive surgery. A validated simulation model of the cryotherapeutic treatment is used to generate a substantial amount of diverse data from different simulation scenarios. We apply machine learning methods on the simulated data to construct a predictive model that provides a prediction for the inner temperature variable based on other system variables whose measurement is more feasible, i.e. skin temperatures. First, we perform feature ranking using the RReliefF method. Next, based on the feature ranking results, we investigate the predictive performance and time/memory efficiency of several predictive modeling methods: linear regression, regression trees, model trees, and ensembles of regression and model trees. Results have shown that using only temperatures from skin sensors as input attributes gives excellent prediction for the temperature in the knee center. Moreover, satisfying predictive accuracy is also achieved using short history of temperatures from just two skin sensors (placed anterior and posterior to the knee) as input variables. The model trees perform the best with prediction error in the same range as the accuracy of the simulated data (0.1°C). Furthermore, they satisfy the requirements for small memory size and real-time response. We successfully validate the best performing model tree with real data from in vivo temperature measurement from a patient undergoing cryotherapy after ACL reconstruction.

Knee temperatures measured in vivo after arthroscopic ACL reconstruction followed by cryotherapy with gel-packs or computer controlled heat extraction.

PURPOSE: To obtain in vivo data about intra- and extra-articular knee temperatures to assess the effectiveness of two cryotherapeutic methods-conventional cooling with gel-packs and computer controlled cryotherapy following anterior cruciate ligament (ACL) reconstructive surgery. METHODS: Twenty patients were arbitrarily assigned for cryotherapy after ACL reconstruction: 8 patients with frozen gel-packs and 12 patients with computer controlled cryotherapy with constant temperatures of the cooling liquid in the knee pads. The treatment was performed for 12 h. Temperatures were measured with two thermo sensors in catheters placed intraarticularly and subcutaneously, four sensors on the skin and one sensor under protective bandage, every second for 16 h after surgery. RESULTS: In the first 2 h of treatment, there were no significant differences (n.s.) between the groups in temperatures in the intracondylar notch. After 4 h of cryotherapy, the temperatures were significantly lower on the skin (24.6 ± 2.8 and 31.4 ± 1.3 °C, p < 0.01) and in the subcutaneous tissue (28.6 ± 5.7 and 34.6 ± 1.4 °C, p = 0.01), and the difference between the temperature in the intracondylar notch and the subcutaneous tissue was significantly greater (4.0 ± 3.0 and 0.8 ± 0.6 °C, p = 0.01) in the computer controlled cryotherapy group compared to the gel-pack group. CONCLUSIONS: The cooling effect of the arthroscopy irrigation fluid on the knee temperature is evident in the first 2 h of treatment. The energy extraction is significantly more effective and controllable by computer controlled cryotherapy than with frozen gel-packs. LEVEL OF EVIDENCE: Prospective comparative study, Level II.

Two proximal skin electrodes--a respiration rate body sensor.

We propose a new body sensor for extracting the respiration rate based on the amplitude changes in the body surface potential differences between two proximal body electrodes. The sensor could be designed as a plaster-like reusable unit that can be easily fixed onto the surface of the body. It could be equipped either with a sufficiently large memory for storing the measured data or with a low-power radio system that can transmit the measured data to a gateway for further processing. We explore the influence of the sensor’s position on the quality of the extracted results using multi-channel ECG measurements and considering all the pairs of two neighboring electrodes as potential respiration-rate sensors. The analysis of the clinical measurements, which also include reference thermistor-based respiration signals, shows that the proposed approach is a viable option for monitoring the respiration frequency and for a rough classification of breathing types. The obtained results were evaluated on a wireless prototype of a respiration body sensor. We indicate the best positions for the respiration body sensor and prove that a single sensor for body surface potential difference on proximal skin electrodes can be used for combined measurements of respiratory and cardiac activities.