Compression of morphologically similar ECG complexes using neural networks

The authors outline a compression approach capable of achieving low bit rates while reconstructing the signal with a high fidelity. The approach does not involve any clinical classification or parameterisation. It can be seen from the results presented that the use of this neural network compression system can provide excellent compression performance, especially where low bit rates and low reconstruction errors are required. Current work is focusing on a full implementation of the compression system, including the precompression classification stage. Particular emphasis is now being placed on the compression of low SNR ECG signals such as those found in a realistic ambulatory recording environment.