Tamara Skoric1, Omer Mohamoud2, Branislav Milovanovic3, Nina Japundzic-Zigon4, Dragana Bajic5. 1. University of Novi Sad Faculty of Technical Sciences, Centre of excellence CEVAS, Trg Dositeja Obradovica 6, 21000 Novi Sad, Serbia. Electronic address: ceranic@uns.ac.rs. 2. University of Novi Sad Faculty of Technical Sciences, Centre of excellence CEVAS, Trg Dositeja Obradovica 6, 21000 Novi Sad, Serbia. Electronic address: hafshook@gmail.com. 3. University of Belgrade, School of Medicine, Laboratory for Cardiovascular, Pharmacology, Serbia; Bezanijska kosa hospital, Neurocardiology department, 11000 Belgrade, Serbia. Electronic address: branislav_milovanovic@vektor.net. 4. University of Belgrade, School of Medicine, Laboratory for Cardiovascular, Pharmacology, Serbia. Electronic address: nzigon@med.bg.ac.rs. 5. University of Novi Sad Faculty of Technical Sciences, Centre of excellence CEVAS, Trg Dositeja Obradovica 6, 21000 Novi Sad, Serbia. Electronic address: dragana.bajic@gmail.com.
Abstract
OBJECTIVES: Personalised monitoring in health applications has been recognised as part of the mobile crowdsensing concept, where subjects equipped with sensors extract information and share them for personal or common benefit. Limited transmission resources impose the use of local analyses methodology, but this approach is incompatible with analytical tools that require stationary and artefact-free data. This paper proposes a computationally efficient binarised cross-approximate entropy, referred to as (X)BinEn, for unsupervised cardiovascular signal processing in environments where energy and processor resources are limited. METHODS: The proposed method is a descendant of the cross-approximate entropy ((X)ApEn). It operates on binary, differentially encoded data series split into m-sized vectors. The Hamming distance is used as a distance measure, while a search for similarities is performed on the vector sets. The procedure is tested on rats under shaker and restraint stress, and compared to the existing (X)ApEn results. RESULTS: The number of processing operations is reduced. (X)BinEn captures entropy changes in a similar manner to (X)ApEn. The coding coarseness yields an adverse effect of reduced sensitivity, but it attenuates parameter inconsistency and binary bias. A special case of (X)BinEn is equivalent to Shannon's entropy. A binary conditional entropy for m =1 vectors is embedded into the (X)BinEn procedure. CONCLUSION: (X)BinEn can be applied to a single time series as an auto-entropy method, or to a pair of time series, as a cross-entropy method. Its low processing requirements makes it suitable for mobile, battery operated, self-attached sensing devices, with limited power and processor resources.
OBJECTIVES: Personalised monitoring in health applications has been recognised as part of the mobile crowdsensing concept, where subjects equipped with sensors extract information and share them for personal or common benefit. Limited transmission resources impose the use of local analyses methodology, but this approach is incompatible with analytical tools that require stationary and artefact-free data. This paper proposes a computationally efficient binarised cross-approximate entropy, referred to as (X)BinEn, for unsupervised cardiovascular signal processing in environments where energy and processor resources are limited. METHODS: The proposed method is a descendant of the cross-approximate entropy ((X)ApEn). It operates on binary, differentially encoded data series split into m-sized vectors. The Hamming distance is used as a distance measure, while a search for similarities is performed on the vector sets. The procedure is tested on rats under shaker and restraint stress, and compared to the existing (X)ApEn results. RESULTS: The number of processing operations is reduced. (X)BinEn captures entropy changes in a similar manner to (X)ApEn. The coding coarseness yields an adverse effect of reduced sensitivity, but it attenuates parameter inconsistency and binary bias. A special case of (X)BinEn is equivalent to Shannon's entropy. A binary conditional entropy for m =1 vectors is embedded into the (X)BinEn procedure. CONCLUSION: (X)BinEn can be applied to a single time series as an auto-entropy method, or to a pair of time series, as a cross-entropy method. Its low processing requirements makes it suitable for mobile, battery operated, self-attached sensing devices, with limited power and processor resources.