Session S62.4

Data Compression for Implantable Medical Devices

LA Koyrakh*

St. Jude Medical
St. Paul, MN, USA

Introduction: Implantable devices have limited memory, computational and battery power resources, while collecting, processing and transmitting out information from potentially many sensors. These limitations require that information within the devices be efficiently compressed. Such data compression presents a challenging task, as it must provide high fidelity of the waveform reproduction and high compression ratios on limited size data frames. Also, it must efficiently run on ultra low power hardware, and allow flexible, yet automatic configuration, based on the type of data to be compressed.
Methods: The compression algorithm was implemented as a bit accurate Matlab simulation, consisting of the following major steps:
1. Integer wavelet transform, which is selectable and could be switched off.
2. Data filtration, performed on bit boundaries, which simplifies ASIC implementation, since no number comparison logic is needed. The filtration thresholds could be made different in different wavelet sub-bands, controlled by a single parameter.
3. Quantization of the filtered data in the time or transform domain, coupled with filtration. Two selectable quantization schemes could be utilized based on the signal properties: linear and deadzone.
4. Original adaptive data encoding. Our approach only requires basic logical operations such as bit counters and shifts, and is highly optimized for implementation in implantable device hardware. For high reliability each compressed data frame contains all information needed for decompression.
Results: The proposed compression algorithm was applied to data from the public PhysioNet database. Frames of 1024 samples of human ECG were digitized at 12 bits, 1000 Hz sampling rate, and compressed with distortions below 2%. The bit compression ratios were 7.89 ± 0.45, achieving above 85% of the entropy limit for each individual data frame. The algorithm showed comparable performance when applied to other waveforms, such as ventricular pressure and EEG.
Conclusions: The compression algorithm is efficient on data collected by implantable devices, and could be used in various applications, helping to reduce memory requirements and the battery energy spent on the information transmission to and from the implantable device.

(Abstract Control Number: 107)