LAUSR

Energy consumption has been a major bottleneck in wireless sensor network applications. This article presents wireless interconnected smart biomedical sensors for cardiac arrhythmia detection and remote monitoring of a group of in-house static patients arranged in a regular grid matrix. In-node continuous detection of arrhythmias is performed using an autoencoder with a decision tree (DT) classifier from single-channel ECG data. The accumulated arrhythmic episodes over some time in each node are delivered through a residual energy-based intelligent routing to a mobile sink, which uploads them to a cloud-based server for remote access by an expert physician. The energy consumption of the node and network is minimized in three stages. First, an event-triggered data routing strategy allows intermittent transmission instead of a continuous one. Second, the arrhythmic episodes are delivered to a mobile sink (a caregiver) based on its location sensing to minimize the number of hops in a routing cone. Third, an adaptive power level adjustment at the destination node is done in accordance with the mobility of the sink. The scheme was hardware implemented with 15 sensor nodes utilizing an ARM-based standalone controller in a 52.68 sqm floor area with a latency of 1.69 ms per byte and energy consumption of 0.338 mJ per byte, with a packet reception accuracy of 97.8% over 4000 sessions of transmission. The blind test accuracy of arrhythmia detection over nearly 20 442 cardiac beats from the MIT BIH arrhythmia dataset containing NSR, L, R, and V beats was 99.48% with a beat abnormality detection latency of 16 ms.