Computerized electrocardiogram (ECG) interpretation plays a critical role in the clinical ECG workflow1. Widely available digital ECG data and the algorithmic paradigm of deep learning2 present an opportunity to substantially improve the accuracy and scalability of automated ECG analysis. However, a comprehensive evaluation of an end-to-end deep learning approach for ECG analysis across a wide variety of diagnostic classes has not been previously reported. Here, we develop a deep neural network (DNN) to classify 12 rhythm classes using 91,232 single-lead ECGs from 53,877 patients who used a single-lead ambulatory ECG monitoring device. When validated against an independent test dataset annotated by a consensus committee of board-certified practicing cardiologists, the DNN achieved an average area under the receiver operating characteristic curve (AUC) of 0.97. The average F1 score, which is the harmonic mean of the positive predictive value and sensitivity, for the DNN (0.837) exceeded that of average cardiologists (0.780). With specificity fixed at the average specificity achieved by cardiologists, the sensitivity of the DNN exceeded the average cardiologist sensitivity for all rhythm classes. These findings demonstrate that an end-to-end deep learning approach can classify a broad range of distinct arrhythmias from single-lead ECGs with high diagnostic performance similar to that of cardiologists. If confirmed in clinical settings, this approach could reduce the rate of misdiagnosed computerized ECG interpretations and improve the efficiency of expert-human ECG interpretation by accurately triaging or prioritizing the most urgent conditions.

Local delivery of drugs to the inner ear has the potential to treat inner ear disorders including permanent hearing loss or deafness. Current mathematical models describing the pharmacokinetics of drug delivery to the inner ear have been based on large rodent studies with invasive measurements of concentration at few locations within the cochlea. Hence, estimates of clearance and diffusion parameters are based on fitting measured data with limited spatial resolution to a model. To overcome these limitations, we developed a noninvasive imaging technique to monitor and characterize drug delivery inside the mouse cochlea using micro-computed tomography (μCT). To increase the measurement accuracy, we performed a subject-atlas image registration to exploit the information readily available in the atlas image of the mouse cochlea and pass segmentation or labeling information from the atlas to our μCT scans. The approach presented here has the potential to quantify concentrations at any point along fluid-filled scalae of the inner ear. This may permit determination of spatially dependent diffusion and clearance parameters for enhanced models.

Abstract-In a network with a high density of wireless nodes, we model flow of information by a continuous vector field known as the information flow vector field. We use a mathematical model that translates a communication network composed of a large but finite number of sensors into a continuum of nodes on which information flow is formulated by a vector field. The magnitude of this vector field is the intensity of the communication activity, and its orientation is the direction in which the traffic is forwarded. The information flow vector field satisfies a set of Neumann boundary conditions and a partial differential equation (PDE) involving the divergence of information, but the divergence constraint and Neumann boundary conditions do not specify the information flow vector field uniquely, and leave us freedom to optimize certain measures within their feasible set. Therefore, we introduce a p-norm flow optimization problem in which we minimize the p-norm of information flow vector field over the area of the network. This problem is a convex optimization problem, and we use sequential quadratic programming (SQP) to solve it. SQP is known for numerical stability and fast convergence to the optimal solution in convex optimization problems. By using standard SQP on p-norm flow optimization, we prove that the solution of each iteration of SQP is uniquely specified by an elliptic PDE with generalized Neumann boundary conditions. The p-norm flow optimization shows interesting properties for different values of p. For example, if p is close to one, the information routes resemble the geometric shortest paths of the sources and sinks, and for p = 2, the information flow shows an analogy to electrostatics. For infinitely large values of p, the problem minimizes the maximum magnitude of the information vector field over the network, and hence it achieves maximum load balancing.

In the version of this article originally published, there was an error in the legend for Extended Data Fig. 7. The legend for panel f was originally: "f, FACS analysis of IL7R-, CD62Land CD45RAexpression on TRAC-1928ζ and TRAC-1XX CAR T cells at day 63 post CAR infusion (representative for at least n = 3 mice per group in one independent experiment). " The legend should have been: "f, FACS analysis of IL7R + , CD62L + and CD45RA + expression on TRAC-1928ζ and TRAC-1XX CAR T cells at day 63 post CAR infusion (representative for at least n = 3 mice per group in one independent experiment). " The error has been corrected in the HTML and PDF versions of this article.

With non-uniform traffic patterns in wireless sensor networks due to the manyto-one nature of communications, the traditional definition of connectivity in graph theory does not seem to be sufficient to satisfy the requirements of sensor networks. In this work, a new notion of connectivity (called pathimplementability) is defined which represents the ability of sensor nodes to relay traffic along a given direction field (referred to as information flow vector field) D. The magnitude of information flow is proportional to the traffic flux (per unit length) passing through any point in the network, and its direction is toward the flow of traffic. The flow field may be obtained from engineering knowledge or as a solution to an optimization problem. In either case, information flow flux lines are abstract paths that are assumed to represent desired paths for flow of traffic. In a sensor network with a given flow field D(x, y), we show that a density of n(x, y) = O(| D(x, y)| 2 ) sensor nodes is not sufficient to implement the flow field. On the other hand, by increasing the density of wireless nodes to n(x, y) = O(| D(x, y)| 2 log | D(x, y)|), the flow field becomes path-implementable. Path-implementability requires more nodes than simple connectivity. However, it guarantees existence of enough paths connecting the information source to the sink so that all the traffic can be transmitted to the sink. We also propose a joint MAC and routing protocol to forward traffic along the flow field; the proposed tier-based scheme can be further exploited to build lightweight protocol stacks which meet the specific requirements of dense sensor networks.