Deep Learning Algorithms to Detect Murmurs Associated With Structural Heart Disease.

Background The success of cardiac auscultation varies widely among medical professionals, which can lead to missed treatments for structural heart disease. Applying machine learning to cardiac auscultation could address this problem, but despite recent interest, few algorithms have been brought to clinical practice. We evaluated a novel suite of Food and Drug Administration-cleared algorithms trained via deep learning on >15 000 heart sound recordings. Methods and Results We validated the algorithms on a data set of 2375 recordings from 615 unique subjects. This data set was collected in real clinical environments using commercially available digital stethoscopes, annotated by board-certified cardiologists, and paired with echocardiograms as the gold standard. To model the algorithm in clinical practice, we compared its performance against 10 clinicians on a subset of the validation database. Our algorithm reliably detected structural murmurs with a sensitivity of 85.6% and specificity of 84.4%. When limiting the analysis to clearly audible murmurs in adults, performance improved to a sensitivity of 97.9% and specificity of 90.6%. The algorithm also reported timing within the cardiac cycle, differentiating between systolic and diastolic murmurs. Despite optimizing acoustics for the clinicians, the algorithm substantially outperformed the clinicians (average clinician accuracy, 77.9%; algorithm accuracy, 84.7%.) Conclusions The algorithms accurately identified murmurs associated with structural heart disease. Our results illustrate a marked contrast between the consistency of the algorithm and the substantial interobserver variability of clinicians. Our results suggest that adopting machine learning algorithms into clinical practice could improve the detection of structural heart disease to facilitate patient care.

Automated detection of low ejection fraction from a one-lead electrocardiogram: application of an AI algorithm to an electrocardiogram-enabled Digital Stethoscope.

Aims: Electrocardiogram (ECG)-enabled stethoscope (ECG-Scope) acquires a single-lead ECGs during cardiac auscultation and may facilitate real-time screening for pathologies not routinely identified by cardiac auscultation alone. We previously demonstrated an artificial intelligence (AI) algorithm can identify left ventricular dysfunction (LVSD) [defined as ejection fraction (EF) ≤ 40%] with an area under the curve (AUC) of 0.91 using a 12-lead ECG. Methods and results: One hundred patients referred for clinically indicated echocardiography were prospectively recruited. ECG-Scope recordings with the patient supine and sitting were obtained in multiple electrode locations at the time of the echocardiogram. The AI algorithm for the detection of LVSD was retrained using single leads from ECG-12 and validated against ECG-Scope to determine accuracy for low EF detection (≤35%, <40%, or <50%). We evaluated the algorithm with respect to body position and lead location. Amongst 100 patients (aged 61.3 ± 13.8; 61% male, BMI: 30.0 ± 5.4), eight had EF≤40%, and six had EF 40-50%. The best single recording position was V2 with the patient supine [AUC: 0.88 (CI: 0.80-0.97) for EF≤35%, 0.85 (CI: 0.75-0.95) for EF≤40%, and 0.81 (CI: 0.71-0.90) for EF < 50%]. When using an AI model to select the recording automatically, AUC was 0.91 (CI: 0.84-0.97) for EF≤35%, 0.89 (CI: 0.83-0.96) for EF≤40%, and 0.84 (CI: 0.73-0.94) for EF < 50%. Conclusion: An AI algorithm applied to an ECG-enabled stethoscope recording in standard auscultation positions reliably detected the presence of a low EF in this prospective study of patients referred for echocardiography. The ability to screen patients with a possible low EF during routine physical examination may facilitate rapid detection of LVSD.

Deep Learning Algorithm for Automated Cardiac Murmur Detection via a Digital Stethoscope Platform.

Background Clinicians vary markedly in their ability to detect murmurs during cardiac auscultation and identify the underlying pathological features. Deep learning approaches have shown promise in medicine by transforming collected data into clinically significant information. The objective of this research is to assess the performance of a deep learning algorithm to detect murmurs and clinically significant valvular heart disease using recordings from a commercial digital stethoscope platform. Methods and Results Using >34 hours of previously acquired and annotated heart sound recordings, we trained a deep neural network to detect murmurs. To test the algorithm, we enrolled 962 patients in a clinical study and collected recordings at the 4 primary auscultation locations. Ground truth was established using patient echocardiograms and annotations by 3 expert cardiologists. Algorithm performance for detecting murmurs has sensitivity and specificity of 76.3% and 91.4%, respectively. By omitting softer murmurs, those with grade 1 intensity, sensitivity increased to 90.0%. Application of the algorithm at the appropriate anatomic auscultation location detected moderate-to-severe or greater aortic stenosis, with sensitivity of 93.2% and specificity of 86.0%, and moderate-to-severe or greater mitral regurgitation, with sensitivity of 66.2% and specificity of 94.6%. Conclusions The deep learning algorithm's ability to detect murmurs and clinically significant aortic stenosis and mitral regurgitation is comparable to expert cardiologists based on the annotated subset of our database. The findings suggest that such algorithms would have utility as front-line clinical support tools to aid clinicians in screening for cardiac murmurs caused by valvular heart disease. Registration URL: https://clinicaltrials.gov; Unique Identifier: NCT03458806.

Dependence of brown adipose tissue function on CD36-mediated coenzyme Q uptake.

Brown adipose tissue (BAT) possesses the inherent ability to dissipate metabolic energy as heat through uncoupled mitochondrial respiration. An essential component of the mitochondrial electron transport chain is coenzyme Q (CoQ). While cells synthesize CoQ mostly endogenously, exogenous supplementation with CoQ has been successful as a therapy for patients with CoQ deficiency. However, which tissues depend on exogenous CoQ uptake as well as the mechanism by which CoQ is taken up by cells and the role of this process in BAT function are not well understood. Here, we report that the scavenger receptor CD36 drives the uptake of CoQ by BAT and is required for normal BAT function. BAT from mice lacking CD36 displays CoQ deficiency, impaired CoQ uptake, hypertrophy, altered lipid metabolism, mitochondrial dysfunction, and defective nonshivering thermogenesis. Together, these data reveal an important new role for the systemic transport of CoQ to BAT and its function in thermogenesis.

Active sensing of target location encoded by cortical microstimulation.

Cortical microstimulation has been proposed as a method to deliver sensory percepts to circumvent damaged sensory receptors or pathways. However, much of perception involves the active movement of sensory organs and the integration of information across sensory and motor modalities. The efficacy of cortical microstimulation in such an active sensing paradigm has not been demonstrated. We report a novel behavioral paradigm which delivers microstimulation in real-time based on a rat's movements and show that rats can perform sensorimotor integration with electrically delivered stimuli. Using a real-time whisker tracking system, we delivered microstimulation in barrel cortex of actively whisking rats when their whisker crossed a particular spatial location which defined the target. Rats learned to integrate microstimulation cues with their knowledge of whisker position to infer target location along the rostro-caudal axis in less than 200 ms. In a separate experiment, we found that rats trained to respond to cortical microstimulation responded similarly to whisker deflections while ignoring auditory distracters, suggesting that barrel cortex stimulation may be perceptually similar to somatosensory stimuli. This ability to deliver sensory percepts using cortical microstimulation in an active sensing system might have significant implications for the development of sensorimotor neuroprostheses.

In vitro and in vivo evaluation of PEDOT microelectrodes for neural stimulation and recording.

Cortical neural prostheses require chronically implanted small-area microelectrode arrays that simultaneously record and stimulate neural activity. It is necessary to develop new materials with low interface impedance and large charge transfer capacity for this application and we explore the use of conducting polymer poly(3,4-ethylenedioxythiophene) (PEDOT) for the same. We subjected PEDOT coated electrodes to voltage cycling between -0.6 and 0.8 V, 24 h continuous biphasic stimulation at 3 mC/cm² and accelerated aging for four weeks. Characterization was performed using cyclic voltammetry, electrochemical impedance spectroscopy, and voltage transient measurements. We found that PEDOT coated electrodes showed a charge injection limit 15 times higher than Platinum Iridium (PtIr) electrodes and electroplated Iridium Oxide (IrOx) electrodes when using constant current stimulation at zero voltage bias. In vivo chronic testing of microelectrode arrays implanted in rat cortex revealed that PEDOT coated electrodes show higher signal-to-noise recordings and superior charge injection compared to PtIr electrodes.

Wireless channel characterization for mm-size neural implants.

This paper discusses an approach to modeling and characterizing wireless channel properties for mm-size neural implants. Full-wave electromagnetic simulation was employed to model signal propagation characteristics in biological materials. Animal tests were carried out, proving the validity of the simulation model over a wide range of frequency from 100MHz to 6GHz. Finally, effects of variability and uncertainty in human anatomy and dielectric properties of tissues on these radio links are explored.

Decoder remapping to counteract neuron loss in brain-machine interfaces.

Variability of single-unit neural recordings can significantly affect the overall performance achieved by brain machine interfaces (BMI). In this paper, we present a novel technique to adapt a linear filter commonly used in BMI to compensate for loss of neurons from the recorded neural ensemble, thus minimizing loss in performance. We simulate the gains achieved by this technique using a model of the learning process during closed-loop BMI operation. This simulation suggests that we can adapt to the loss of 24% of the neurons controlling a BMI with only 13% drop in performance.

Investigating neural correlates of behavior in freely behaving rodents using inertial sensors.

Simultaneous behavior and multielectrode neural recordings in freely behaving rodents holds great promise to study the neural bases of behavior and disease models in combination with genetic manipulations. Here, we introduce the use of three-axis accelerometers to characterize the behavior of rats and mice during chronic neural recordings. These sensors were small and light enough to be worn by rodents and were used to record three-axis acceleration during freely moving behavior. A two-layer neural network-based pattern recognition algorithm was developed to extract the natural behavior of mice from the acceleration data. Successful recognition of resting, eating, grooming, and rearing are shown using this approach. The inertial sensors were combined with continuous 24-h recordings of neural data from the striatum of mice to characterize variations in neural activity with circadian cycles and to study the neural correlates of spontaneous action initiation. Finally, accelerometers were used to study the performance of rodents in traditional operant conditioning, where they were used to extract the reaction time of rodents. Thus the addition of accelerometer recordings of rodents to chronic multielectrode neural recordings provides great value for a number of neuroscience applications.

Exploiting the 1/f structure of neural signals for the design of integrated neural amplifiers.

Neural amplifiers require a large time-constant high-pass filter at approximately 1Hz to reject large DC offsets while amplifying low frequency neural signals. This high pass filter is typically realized using large area capacitors and teraohm resistances which makes integration difficult. In this paper, we present a novel topology for a neural amplifier which exploits the (1/f)(n) power spectra of local field potentials (LFP). Using a high-pass filter at approximately 100Hz, we pre-filter the LFP before amplification. Post digitization, we can recover the LFP signal by building the inverse of the high pass filter in software. We built an array of neural amplifiers based on this principle and tested it on rats chronically implanted with microelectrode arrays. We found that we could recover the initial LFP signal and the power spectral information over time with correlation coefficient greater than 0.94.

Behavioral modulation of stimulus-evoked oscillations in barrel cortex of alert rats.

Stimulus-evoked oscillations have been observed in the visual, auditory, olfactory and somatosensory systems. To further our understanding of these oscillations, it is essential to study their occurrence and behavioral modulation in alert, awake animals. Here we show that microstimulation in barrel cortex of alert rats evokes 15-18 Hz oscillations that are strongly modulated by motor behavior. In freely whisking rats, we found that the power of the microstimulation-evoked oscillation in the local field potential was inversely correlated to the strength of whisking. This relationship was also present in rats performing a stimulus detection task suggesting that the effect was not due to sleep or drowsiness. Further, we present a computational model of the thalamocortical loop which recreates the observed phenomenon and predicts some of its underlying causes. These findings demonstrate that stimulus-evoked oscillations are strongly influenced by motor modulation of afferent somatosensory circuits.

A system for neural recording and closed-loop intracortical microstimulation in awake rodents.

There is growing interest in intracortical microstimulation as a means of providing sensory input in neuroprosthetic systems. We believe that precisely controlling the timing and parameters of stimulation in closed loop can significantly improve the efficacy of this technique. Here, we present a system for closed-loop microstimulation in awake rodents chronically implanted with multielectrode arrays. The system interfaces with existing commercial recording and stimulating hardware. Using custom-made hardware, we can stimulate and record from electrodes on the same implanted array and significantly reduce the stimulation artifact. Stimulation sequences can either be preprogrammed or triggered by neural or behavioral events. Specifically, this system can provide feedback stimulation in response to action potentials or features in the local field potential recorded on any of the electrodes within 15 ms. It can also trigger stimulation based on behavioral events, such as real-time tracking of rat whiskers captured with high-speed video. We believe that this system, which can be recreated easily, will help to significantly refine the technique of intracortical microstimulation and advance the field of neuroprostheses.

Wireless inertial sensors for monitoring animal behavior.

Wireless sensors were designed which are small and light enough to be worn by small animals such as rats. These sensors are used to record three axes acceleration data from animals during natural behavior in a cage. The behavior of the animal is further extracted from the recorded acceleration data using neural network based pattern recognition algorithms. Successful recognition of eating, grooming and standing are demonstrated using this approach. Finally another potential application of this research is demonstrated in behavioral neuroscience by showing correlations between action potentials recorded from the motor cortex of a rat and acceleration data.